
At present, special efforts are being made to develop the strategies allowing for activation of long-lasting antigen-specific immune tolerance in therapy of allergic and autoimmune diseases. Some of these therapeutic approaches are aimed at modulating cell functions at genetic level by using miRNA-based and miRNA-targeting treatments. Simultaneously, the crucial role of extracellular vesicles as natural miRNA conveyors is highlighted for induction of antigen-specific immune tolerance, especially that they appear to be easily manipulatable for therapeutic applications. Among other immune-related miRNAs, miR-150 is getting special attention as it is differently expressed by immune cells at various stages of their maturation and differentiation. In addition, miR-150 is involved in different signaling cascades orchestrating humoral and cell-mediated mechanisms of both innate and adaptive immune responses. Therefore, miR-150 is considered a master regulator of immunity in mammals. Currently, physiological miR-150-dependent regulatory circuits and causes of their malfunctioning that underlie the pathogenesis of allergic and autoimmune disorders are being unraveled. Thus, present review summarizes the current knowledge of the role of miR-150 in the pathogenesis and complications of these diseases. Furthermore, the involvement of miR-150 in regulation of immune responses to allergens and self-antigens and in induction of antigen-specific immune tolerance is discussed with the special emphasis on the therapeutic potential of this miRNA.
Citation: Katarzyna Nazimek. The complex functions of microRNA-150 in allergy, autoimmunity and immune tolerance[J]. AIMS Allergy and Immunology, 2021, 5(4): 195-221. doi: 10.3934/Allergy.2021016
[1] | Andrea Borio, Martina Busetto, Francesca Marcon . SUPG-stabilized stabilization-free VEM: a numerical investigation. Mathematics in Engineering, 2024, 6(1): 173-191. doi: 10.3934/mine.2024008 |
[2] | Siddhartha Mishra . A machine learning framework for data driven acceleration of computations of differential equations. Mathematics in Engineering, 2019, 1(1): 118-146. doi: 10.3934/Mine.2018.1.118 |
[3] | Daniele Cerroni, Florin Adrian Radu, Paolo Zunino . Numerical solvers for a poromechanic problem with a moving boundary. Mathematics in Engineering, 2019, 1(4): 824-848. doi: 10.3934/mine.2019.4.824 |
[4] | Yu Leng, Lampros Svolos, Dibyendu Adak, Ismael Boureima, Gianmarco Manzini, Hashem Mourad, Jeeyeon Plohr . A guide to the design of the virtual element methods for second- and fourth-order partial differential equations. Mathematics in Engineering, 2023, 5(6): 1-22. doi: 10.3934/mine.2023100 |
[5] | Paola F. Antonietti, Ilario Mazzieri, Laura Melas, Roberto Paolucci, Alfio Quarteroni, Chiara Smerzini, Marco Stupazzini . Three-dimensional physics-based earthquake ground motion simulations for seismic risk assessment in densely populated urban areas. Mathematics in Engineering, 2021, 3(2): 1-31. doi: 10.3934/mine.2021012 |
[6] | Andrea Manzoni, Alfio Quarteroni, Sandro Salsa . A saddle point approach to an optimal boundary control problem for steady Navier-Stokes equations. Mathematics in Engineering, 2019, 1(2): 252-280. doi: 10.3934/mine.2019.2.252 |
[7] | Jinghong Li, Hongyu Liu, Wing-Yan Tsui, Xianchao Wang . An inverse scattering approach for geometric body generation: a machine learning perspective. Mathematics in Engineering, 2019, 1(4): 800-823. doi: 10.3934/mine.2019.4.800 |
[8] | Giorgia Franchini, Valeria Ruggiero, Federica Porta, Luca Zanni . Neural architecture search via standard machine learning methodologies. Mathematics in Engineering, 2023, 5(1): 1-21. doi: 10.3934/mine.2023012 |
[9] | Thomas J. Radley, Paul Houston, Matthew E. Hubbard . Quadrature-free polytopic discontinuous Galerkin methods for transport problems. Mathematics in Engineering, 2024, 6(1): 192-220. doi: 10.3934/mine.2024009 |
[10] | Claudio Canuto, Davide Fassino . Higher-order adaptive virtual element methods with contraction properties. Mathematics in Engineering, 2023, 5(6): 1-33. doi: 10.3934/mine.2023101 |
At present, special efforts are being made to develop the strategies allowing for activation of long-lasting antigen-specific immune tolerance in therapy of allergic and autoimmune diseases. Some of these therapeutic approaches are aimed at modulating cell functions at genetic level by using miRNA-based and miRNA-targeting treatments. Simultaneously, the crucial role of extracellular vesicles as natural miRNA conveyors is highlighted for induction of antigen-specific immune tolerance, especially that they appear to be easily manipulatable for therapeutic applications. Among other immune-related miRNAs, miR-150 is getting special attention as it is differently expressed by immune cells at various stages of their maturation and differentiation. In addition, miR-150 is involved in different signaling cascades orchestrating humoral and cell-mediated mechanisms of both innate and adaptive immune responses. Therefore, miR-150 is considered a master regulator of immunity in mammals. Currently, physiological miR-150-dependent regulatory circuits and causes of their malfunctioning that underlie the pathogenesis of allergic and autoimmune disorders are being unraveled. Thus, present review summarizes the current knowledge of the role of miR-150 in the pathogenesis and complications of these diseases. Furthermore, the involvement of miR-150 in regulation of immune responses to allergens and self-antigens and in induction of antigen-specific immune tolerance is discussed with the special emphasis on the therapeutic potential of this miRNA.
The Galerkin-finite element (FE) method applied to partial differential equations (PDEs) with advection terms dominating over diffusion ones may suffer of numerical instability [22,34]. These numerical instabilities cause the numerical solution to exhibit oscillations that increase in amplitude with a local increment of the transport dominance over diffusion, i.e., as soon as the local Péclet number becomes larger than one. In order to eliminate (or at least mitigate) numerical instabilities, a general employed strategy is to consider the generalized Galerkin method, i.e., the Galerkin method with additional stabilization terms [34]. Examples of stabilization methods to reduce numerical oscillations in the advection dominated regimes are for instance the upwind and streamline-diffusion method, both methods that are not strongly consistent. Instead, examples of strongly consistent methods are the Galerkin-least-squares, the streamline upwind Petrov-Galerkin (SUPG), and the Douglas-Wang methods [6,34,41]. In particular, for these strongly consistent stabilization methods, the formulation depends on the residual of the PDE (in strong formulation), other than on a parameter – called stabilization parameter – whose definition is crucial for the success of the stabilization strategies. Determining the values of such stabilization parameter is not straightforward, especially for 2D and 3D problems. In addition, a universal formulation of such parameter is lacking, especially as it is strongly dependent on data of the model and the numerical setting used for the FE approximation of the PDEs, e.g., low and high order FE methods, spectral element methods, isogeometric methods, etc. [7,12,42]. Some of the formulations for the stabilization parameter have been derived from analytical considerations on the differential problem in 1D, others are suitable only for a specific numerical approximation method, while the most comprehensive ones incorporate some (empirical) dependence on numerical setting, as e.g., the order of the FE [5,6,8,9,10,16,18,38,43,44,45].
In this work, we propose using machine learning (ML) [19,30,50] to make computers learn autonomously the values of the stabilization parameter for the FE approximation of advection dominated PDEs. Artificial neural networks (ANNs) [27,46] are widely popular in ML and Deep Learning for a wide array of applications: they are indeed very versatile tools that are increasingly finding their way in scientific computing [29,31], especially in the context of numerical approximation of PDEs. For instance, as substitute to standard numerical methods, ANNs can be employed as meshless methods in physics informed neural networks (PINNs) to directly approximate the solution of the PDE as it is trained by minimizing the (strong) residual of the PDE [35,36,37]. ANNs are also largely employed in a data-driven fashion in the context of model order reduction for parametric PDEs [17,20,21,40,51] and to enhance the stability properties of numerical methods for PDEs [13]. In fluid dynamics modelling, ANNs are massively adopted for flow features extractions, modelling, optimization and control. For flow features extractions, ANNs are used through clustering and classification to classify wake topologies [11]; for modeling fluid dynamics by reconstructing specific flows such as the near wall field in a turbulent flow [28] and for flow optimization [32], and control for aerodynamics applications [4]. ANNs are also used in a data-driven framework as a manner for providing alternative closure models for stress tensor [14] in Reynolds-Average Navier-Stokes (RANS) equations, for sub-grid scale models in large eddy simulation (LES) turbulence models [23,24,48,49,52], or for model learning input-output relationships in complex physical processes [39].
In this work, we use ANNs to learn the optimal stabilization parameter in advection dominated PDEs that are discretized by means of the FE method: the goal is to enhance the accuracy of the SUPG FE method by optimally selecting the stabilization parameter under different data of the PDE and numerical settings of the FE method. We found that the proposed ANN-enhanced stabilization method allows to improve accuracy and stabilization properties of the numerical solution compared to those results obtained by analytical expressions of the stabilization parameter. The numerical results obtained shed light also on the possibility to apply the presented strategy to learn closure laws for stabilization and turbulence models of fluid dynamics, as for instance to learn the stabilization parameters in the Variational Multiscale–LES model to model transitional and turbulent flows [3,15,54].
This work is organized as follows: in Section 2, we recall the SUPG stabilization method for advection-diffusion equations; in Section 3, we present our numerical strategy to compute an optimal SUPG stabilization parameter through a feed-forward fully connected ANN. In Section 4 we validate our method by comparing the ANN results with those obtained with the 1D advection-diffusion problem from which the expression of the theoretical stabilization parameter has been derived. In Section 5 we first show the ANN's training and we present our numerical results on the 2D advection-diffusion problem used for training and we finally generalize our findings using the ANN's prediction on a different advection-diffusion problem. Finally, in Section 7 we draw our conclusions highlighting possible future developments.
We briefly recall the advection-diffusion differential problem and the SUPG stabilization method for the advection dominated regime.
Let Ω∈Rd,d=1,2,3 be the physical domain with ∂Ω being its boundary. We consider the following problem in the unknown function u:
{−∇⋅(μ∇u)+β⋅∇u=finΩ,u=gon∂Ω, | (2.1) |
where μ, β, and f are assigned functions or constants, with μ∈L∞(Ω), β∈[L∞(Ω)]d, with ∇⋅β∈L2(Ω), and f∈L2(Ω). The Dirichlet datum on the boundary is g∈H1/2(∂Ω). Let Vg={v∈H1(Ω):v|∂Ω=g} and V0=H10(Ω); the weak formulation of Eq (2.1) reads
findu∈Vg : a(u,v)=F(v)for all v∈V0. | (2.2) |
with the bilinear form a(u,v) and the linear functional F(v) respectively defined as:
a(u,v):=∫Ωμ∇u⋅∇vdΩ+∫Ωvβ⋅∇udΩ,F(v):=∫ΩfvdΩ. |
We consider a family of function spaces Vh⊂V (either for Vg and V0) dependent on a parameter h such that dim(Vh)=Nh<∞. Let Xhr={vh∈C0(¯Ω):vh|K∈Pr,for all K∈Th} be the function space of the FE discretization with piecewise Lagrange polynomials of degree r≥1, Th a triangulation of Ω and h the characteristic size of the mesh, comprised of elements K∈Th. By setting Vh=Xrh⋂V0, the Galerkin FE method applied to Eq (2.2) reads
finduh∈Vg,h : a(uh,vh)=F(vh)for all vh∈Vh, | (2.3) |
where Vg,h=Xrh⋂Vg. The standard Galerkin-FE method, in Eq (2.3), can generate numerical oscillations on uh if the problem is dominated by the advection term. In particular, these numerical instabilities can arise if the local Péclet number is Peh>1, where
Peh=|β|h2μ. | (2.4) |
The generalized Galerkin method ([33]) allows for eliminating or mitigating numerical oscillations by adding stabilization terms to the standard Galerkin formulation as
finduh∈Vg,h : a(uh,vh)+bh(uh,vh)=F(vh)for all vh∈Vh. | (2.5) |
In the SUPG method, the additional stabilization term reads:
bh(uh,vh)=∑K∈Th∫KR(uh)τK12(∇⋅(βvh)+β⋅∇vh)dΩ, | (2.6) |
where R(uh) is the residual in strong formulation of Eq (2.1), which is defined as:
R(uh):=−∇⋅(μ∇uh)+β⋅∇uh−f. |
The term τK, appearing in Eq (2.6), is the stabilization parameter, which is the focus of this work. The stabilization parameter τK is generally defined locally, i.e., mesh element by element. In this paper, we consider a uniform stabilization parameter, thus τK=τ for all K∈Th.
A universal and optimal definition of τ in terms of the problem data and numerical settings like the mesh size and FE degree is lacking. An extensive review of stabilization parameters for the SUPG method is reported in [25]. The most common choices for τ come from an analytic derivation made for the advection-diffusion problem in 1D with f=0 and approximated by means of linear finite elements, which reads:
˜τ1=h2|β|ξ(Peh), | (2.7) |
where ξ(θ) is the upwind function:
ξ(θ)=coth(θ)−1θ,θ>0. |
If a uniform mesh is used, as we consider in this paper, then the value of h is uniform over Th; if in addition β and μ are constant, then this implies that τ is uniform over Th. The choice of τ made in Eq (2.7) represents an optimal choice of the stabilization parameter as it yields a nodally exact numerical solution for the 1D advection-diffusion problem if f=0 and the FE polynomial order is r=1 [18,34]. Thus, the stabilization parameter of Eq (2.7) may not be fully effective to provide the optimal stabilization for a general advection-diffusion problem, e.g., to guarantee a nodally exact solution in 2D/3D or when using FE degrees larger than r=1. A commonly used generalization of the formula of Eq (2.7) to higher FE degrees (r>1) is presented in [18] and reads:
˜τr=h2|β|rξ(Pehr). | (2.8) |
Differently from Eq (2.7), the stabilization parameter ˜τr in Eq (2.8) takes into account the contribution of higher polynomial degrees r. Still, this formula is not optimal as it does not guarantee a nodally exact numerical solution. Our goal is to find a general and optimal expression of the stabilization parameter holding for advection-diffusion problems and FE approximations of degree r≥1 to be dependent on: the dimension d, the FE degree r, the mesh size h, the forcing term f, the diffusion coefficient μ and the transport coefficient β.
We present our approach to determine the optimal stabilization parameter τ for the SUPG stabilization method by using an ANN. We consider as features (inputs) of our ANN: the FE degree r, the mesh size h and the global Péclet number Peg:=|β|L/(2μ) of the advection-diffusion problem, where L is the characteristic length of the problem, β∈Rd and μ∈R. As we consider β to be uniform and fixed, using Peg as input corresponds to varying the value of the diffusion coefficient μ. The features (input) of the ANN read:
x(i)=[r,h,Peg]; | (3.1) |
the target (output) of interest is the optimal SUPG stabilization parameter that we denote with τ∗:
y(i)=[τ∗]. | (3.2) |
The first step consist in generating the dataset, i.e., pairs of inputs and outputs to be used for training the ANN (data generation step). This consists in choosing repeatedly and randomly values of the features x(i) in given ranges. These features are used to feed an optimization problem that, through an optimizer and a suitable error measure, provide an optimal stabilization parameter y(i), i.e., the target of the given feature. Such optimization problem considers as error measure E(τ): the mismatch between the numerical solution uh and the exact one uex over the nodes of the FE mesh. E(τ) reads as:
E(τ)=Kh∑k=1|e(xk,τ)|,e(xk,τ)=uh(xk;τ)−u(xk). | (3.3) |
being xk the k-th node of the FE mesh Th and Kh the number of nodes. E(τ) is an approximation of the L1(Ω) norm. To find the minimum of E(τ) for different problem configurations we solve, for each instance of the input parameters x(i), the following optimization problem:
findτ∗ :minτE(τ). | (3.4) |
Thus, the dataset generated consists of m pairs of inputs-outputs (x(i),y(i)), with i=1,…,m. The latter is used for training the ANN. The latter will be then used to predict the optimal stabilization parameter y(j)=τANN to be used for the FE approximation of the advection-diffusion problem in the settings provided by x(j).
We report in Figure 1 a sketch of the procedure used to build the ANN for the prediction of the optimal stabilization parameter for any new feature x(j).
In this section, we validate the proposed numerical strategy by means of a 1D advection-diffusion problem from which the expression of ˜τ1 in Eq (2.7) has been derived. In particular, we consider the advection-diffusion problem in Eq (2.1) with Ω=(0,1), f=0 and with Dirichlet BCs u=0 on x=0 and u=1 on x=1. It admits the following exact solution:
u(x)=exp(βμx)−1exp(βμ)−1. |
We recall that the stabilization parameter ˜τ1 of Eq (2.7) yields a nodally exact numerical solution if Eq (2.1) is solved by linear FE (r=1).
We plot in Figure 2 the error measure E(τ) against the value of the stabilization parameter τ, with h=1/20, Peh=12.5 and by using r=1 (Figure 2a) and r=3 (Figure 2b). We observe that E(τ) shows a minimum in τ∗, thus suggesting the possibility to use an optimization algorithm to solve the problem. Specifically, we employ the L-BFGS-B optimization algorithm from SciPy, an open source library for Python [47]. Instead, the advection-diffusion problem has been solved using the FE open source library FEniCS [2], by applying the SUPG method on a uniform mesh. Particularly, the optimal value τ∗ found for the case r=1 corresponds to the one provided by the theory in Eq (2.7). For the case r=3, we observe a optimal value of τ∗ for which the error is minimized: in this case, it does not corresponds to the stabilization parameter ˜τr provided by the theory in Eq (2.8): as a matter of fact, the latter arises from empirical considerations to extend the parameter ˜τ1 in Eq (2.7) to polynomials degrees r>1. However, this does not ensure a nodally exact numerical solution.
We solve the optimization problem for different values of the local Péclet number 1≤Peh≤250 and we plot in Figure 3 the optimal stabilization parameter τ∗ against the local Péclet number for r=1,2, and 3.
Figure 3a shows that, by employing a linear FE space r=1, the optimal stabilization parameter τ∗ obtained by minimizing E(τ) exactly matches the theoretical one ˜τ1 for every value of Peh, thus confirming the findings of Figure 2a. This also serves as validation of the optimization procedure that we proposed. By recalling that ˜τ1 of Eq (2.7) guarantees the numerical solution to be exact at nodes for this particular advection-diffusion problem and r=1, we can infer that the optimization procedure is meaningful and it can be therefore exploited further, for example for r>1. On the other hand, Figure 3b and 3c show instead a mismatch between the theoretical ˜τr of Eq (2.8) and the optimal one τ∗, thus confirming the findings reported in Figure 2b.
In order to better appreciate the differences in the numerical solutions due to the choice of the stabilization parameters, we report, in Figure 4, a comparison of the error E(τ) obtained by means of the optimal τ∗ and theoretical ˜τr stabilization parameters for r=2 and 3. Moreover, we report in Figure 5 a comparison among the exact solution u, the SUPG-stabilized numerical solution u∗h obtained with the optimal stabilization parameter τ∗, and the numerical solution ˜uh obtained with theoretical one ˜τr. Specifically, we consider FE of degree r=3 and two different values of Peh. In both these cases, the optimal parameter τ∗ leads to a more accurate solution with respect to using the theoretical parameter ˜τr. In particular, when Peh is "small", ˜τr leads to overshooting in the numerical solution, while if Peh is "large" the theoretical stabilization parameter leads to undershooting. Conversely, the optimal parameter τ∗ accurately intercepts the exact solution at the nodes.
We first introduce the training set of our problem and we detail the setup of the ANN that we use in this work. Then, we show the prediction of the stabilization parameter by means of the ANN on different advection-diffusion problems.
We apply the strategy presented so far to a 2D advection-diffusion problem to generate the dataset for the training of the ANN. Specifically, we consider in Eq (2.1): Ω=(0,1)2, f=0, and β=(1,1). We prescribe the following exact solution on the whole boundary ∂Ω:
u(x,y)=e(x/μ)−1e(1/μ)−1+e(y/μ)−1e(1/μ)−1. | (5.1) |
We generate in Ω a structured mesh Th of triangles with FEniCS [2], as shown in Figure 6. We generate the dataset by repeatedly solving the optimization problem of Eq (3.4) for varying set of features as described in Section 3; specifically, we choose the features as reported in Eq (3.1). The complete dataset contains m=900 examples with r={1,2,3}, h={√210,√220,√240} and values of Peg randomly chosend in the range [7,70′710] (uniform distribution), which yields μ∈[10−5,10−1]. We summarize the details for generating the dataset in Table 1. Figure 7 provides an overview of the dataset used for the training of the ANN, specifically the features x(i)=[r,h,Peg] and the target y(i)=[τ∗]. In this case, the characteristic length L is fixed and set equal to L=1.
# Data set (m) | # Training set | # Validation set | r | h | Peg |
900 | 720(80%) | 180(20%) | {1,2,3} | {√210,√220,√240} | randomly in [7,70′710]. |
We train a fully-connected feed-forward ANN on the generated dataset by using the open source library Keras [26] built on top of TensorFlow [1]. We divide the dataset into two parts: a training dataset that takes 80% of the examples to be used for the ANN training and a validation dataset that takes the remaining 20%. We choose the loss function as the mean squared error that measures, for each training feature, the squared mismatch between the prediction of the ANN ˆy(j) and the actual target y(j). Specifically, the loss function is defined as
J=12mm∑j=1(ˆy(j)−y(j))2. | (5.2) |
We normalize the features by subtracting their sample mean and dividing by their sample standard deviation in order to help the weights to better adapt to the different scales of the features. Moreover, the targets and the feature Peg need special care as they are distributed over a wide range of values. Thus, we normalize them using by applying a base 10 logarithm. The normalized features and targets read:
˜x=[r−¯rσr,h−¯hσh,log10(Peg)−¯log10(Peg)σlog10(Peg)],˜y=[−log10(τ∗)], |
where ¯r, ¯h, ¯log10(Peg) are the sample mean of the training features r, h and the logarithm of Peg respectively, while σr, σh and σlog10(Peg) are their sample standard deviations.
In order to find the best performing ANN architecture, we carried out a study by testing different numbers of hidden layers, nodes per layer, optimization algorithm, its learning rate, and batch size. A comparison of the loss function with different architectures is given in Figure 8. Using 3 hidden layers is beneficial in terms of validation error drops, while using more than 64 nodes per layer does not bring to considerable advantages. Finally, we choose an ANN with 3 hidden layers, 64 nodes per layer and an output layer with a single node, all using a rectified linear unit (ReLU) activation function. We display the ANN architecture in Figure 9. We trained the ANN with the SGD optimization algorithm, a constant learning rate of 0.01, and mini-batch size of 32 samples. Moreover, we employed a momentum of 0.9 in the optimization algorithm to update the weights. The trained ANN is available in the GitLab repository [53].
Regarding the computational efficiency of the proposed strategy, we stress the clear distinction between the offline phase (dataset generation and ANN's training) and the online phase, where we use the ANN to predict a new stabilization parameter – alongside the use of the FE solver – for unseen input parameter values. The most demanding part of the strategy is the dataset generation, requiring approximately 2 h on a standard laptop to repeatedly solve the optimization problem 900 times. Moreover, the ANN's training phase required approximately 10'. The testing (online) phase, which is the one that is performed for application purposes, is considerably inexpensive, requiring only the real-time evaluation of a composition of linear functions, comparable with the evaluation of the empirical relation that brings to the theoretical values (few milliseconds).
Now, we compare the predictions of the ANN with the theoretical stabilization parameter ˜τr of Eq (2.8) [18,34]. We show in Figure 10 (left) the stabilization parameter τANN predicted by the ANN by varying mesh size h and global Péclet number Peg. For comparison, we report in Figure 10 (right) the corresponding theoretical stabilization parameter ˜τr. We observe that the overall behavior of the trained ANN's predictions are qualitatively similar of the theoretical stabilization parameter ˜τr. Nevertheless, it can be inferred that the values of the stabilization parameters are almost completely matched with linear FE (r=1), while they quantitatively differ for r=2 and r=3. This was expected as ˜τr for r>1 is an empirical extension of the formula for the case r=1.
The ANN allows to make predictions with features outside the range of values for which it has been trained, that is for unseen values of such features. With this aim, we report in Figure 11 the comparison of the ANN's predictions τANN with ˜τr for the FE degree r=4. This comparison shows a clear difference between the theoretical and ANN's τ for r=4 even though the general trend is maintained similar.
We compare the numerical solutions uANNh and ˜uh obtained by means of the SUPG stabilization method with the parameter τANN predicted by the ANN and the theoretical one ˜τr, respectively; the comparison also involves the exact solution u (5.1) of the 2D advection-diffusion problem used for the training of the ANN. In particular, we display the comparison of the former 2D solutions in Figure 12 along the line (1−h,y) for any y∈[0,1] with: (a) Peh=2, r=1, and h=√2/10 (Figure 12a); (b) Peh=500, r=3, and and h=√2/20 (Figure 12b). We notice that in both the cases, the numerical solution uANNh involving the stabilization parameter τANN provides more accurate results than with the theoretical stabilization paramter ˜τr. In particular, in the case (a), ˜uh involves a much smoother boundary layer and overshoots the exact solutions u, conversely to the nearly nodally exact numerical solution uANNh. In the case (b), uANNh provides a much better representation of the bundary layer, without the undershooting of the solution u exhibited by ˜uh. Furthermore, we report in Table 2 absolute errors between the numerical solutions (uANNh and ˜uh) and the exact one u. The errors, computed in L2(Ω) and H1(Ω) norms, show that the ANN-based SUPG stabilization method very often produces more accurate results than the ones obtained with theoretical stabilization parameter ˜τr, especially for r=3.
˜τr (e=˜uh−u) | τANN (e=uANNh−u) | |||||
r | h | Peh | ||e||L2(Ω) | ||e||H1(Ω) | ||e||L2(Ω) | ||e||H1(Ω) |
1 | √2/10 | 2 | 9.56⋅10−2 | 7.93⋅10−1 | 6.49⋅10−2 | 5.31⋅10−1 |
2 | √2/10 | 2 | 6.48⋅10−2 | 5.31⋅10−1 | 6.50⋅10−2 | 5.33⋅10−1 |
3 | √2/10 | 2 | 6.50⋅10−2 | 5.33⋅10−1 | 6.49⋅10−2 | 5.32⋅10−1 |
1 | √2/20 | 500 | 2.99⋅10−4 | 5.30⋅10−3 | 3.11⋅10−5 | 5.50⋅10−4 |
2 | √2/20 | 500 | 2.71⋅10−2 | 4.86⋅10−1 | 2.50⋅10−2 | 4.48⋅10−1 |
3 | √2/20 | 500 | 1.05⋅10−2 | 1.90⋅10−1 | 1.68⋅10−3 | 3.46⋅10−2 |
Moreover, to assess the ability of the ANN to predict the stabilization parameter out of the training range, we compare in Figure 13 the numerical solutions for high Péclet in the case r=4. We observe that uANNh is more accurate than ˜uh even with FE degree out of the training range.
We check the model generalization of the ANN trained for the 2D advection-diffusion problem with exact solution u of Eq (5.1) by predicting τANN for an unseen 2D advection-diffusion problem. In particular, we consider the advection-diffusion problem of Eq (2.1) in Ω=(0,1)2, with f=1 and β=(1,1). We prescribe the following exact solution on the whole boundary ∂Ω:
u(x,y)=12(x+y)+1−12(e(x/μ)+e(y/μ))e(1/μ)−1. | (5.3) |
We compare in Figure 14 the numerical solutions uANNh and ˜uh with the novel exact solution u for different Péclet and mesh sizes. As for the previous numerical tests, the stabilization parameter τANN predicted by the network provides more accurate numerical solutions uANNh than with the theoretical stabilization parameter ˜τr. In particular, the boundary layers and overall solution behaviours are better represented.
We better assess the former qualitative consideration, by computing the error E(τ) associated with the numerical solutions with SUPG stabilization for the parameters τANN and ˜τr; different values of Peg and r are considered. In particular, Figure 15 compares E(τ) against Peg (in logarithmic scale) for different values of r: the error obtained by using τANN is always lower than the one achieved with ˜τr. These results indicate that τANN, although trained on a specific advection-diffusion problem, can be used to make predictions of the optimal τ for an unseen advection-diffusion problem in place of the theoretical stabilization parameter.
In this section, we consider an advection-diffusion problem with constant forcing term f=1, advection coefficient β=(1,1), and boundary conditions u=0 on ∂Ω. Although there is not an exact solution to the given problem, we consider as "ground truth" (reference) solution a numerical solution obtained on a much finer grid (h=√2/400) without stabilization. In Figure 16, we compare the SUPG numerical solutions ˜uh, uANNh against our "ground truth" solution for two different Péclet numbers. The numerical solution with the stabilization parameter τANN provides more accurate results with respect to the the solution obtained with the theoretical parameter. In particular, we highlight the largest discrepancies observed in the corners of the extracted solution.
In this section, we cope the case of a non costant forcing term by considering advection-diffusion problem of Eq (2.1) in Ω=(0,1)2 with β=(1,1), and we prescribe the following exact solution on the whole boundary ∂Ω:
u(x,y)=−atan((x−1/2)2+(y−1/2)2−1/16)√μ. | (5.4) |
We display the exact solution of the considered problem in Figure 17.
In particular, Figures 18 and 19 show E(τ) against Peg for different values of r with two different meshes with h=√2/10 and h=√2/20, respectively. The error obtained by using τANN is always comparable to the one achieved with ˜τr, for both mesh levels and for all the FE degrees considered. These results suggest that τANN, although trained on a specific advection-diffusion problem with f=0, is robust with respect to unseen parameters and data in the advection-diffusion problem, including non constant f. Nevertheless, further studies can be conducted to better assess the role of a non-constant forcing term into the ANN's training phase, possibly encompassing the accuracy of the theoretical stabilization parameter in this scenario too.
We investigate the effect of the advection direction on the value of the stabilization parameter. We define θ as the angle between the advection velocity and the x-axis: β=(|β|cosθ,|β|sinθ)T. We carry out the optimization strategy introduced in Section 3 including also θ in the input parameter x(i). We apply the optimization to the advection-diffusion problem with exact solution in Eq (5.3) varying θ in [π/12,π/2] and we report a comparison between the optimal, theoretical and ANN's stabilization parameter in Figure 20 (left). We set |β|=√2, Peg=7′071, h=√2/20, r=3. We recall that, differently from the optimization strategy, in this plot we are still employing the ANN trained without accounting for the advection direction. We can observe that the optimal τ is affected by θ, a feature that is not accounted by the theoretical stabilization parameter. Furthermore, in Figure 20 (right), we compare the mean errors obtained with the optimal, theoretical and ANN's stabilization parameter. It can be observed that the ANN still provides an advantage with respect to the theoretical stabilization. However, we believe that by including the advection direction as an additional feature in the training phase of the ANN, a higher accuracy can be obtained that the one achieved in this paper, where a single direction (θ=π4) has been considered.
In this work, we presented an approach based on machine learning and ANN to compute the optimal stabilization parameter to be used in the SUPG FE approximation of advection-diffusion problems. Indeed, albeit the expression of the stabilization parameter is available for 1D problems and FE of degree r=1, its extension to more general advection-diffusion problems (2D and 3D) and FE degrees r>1 is still lacking.
We validated our approach against the 1D case, for which the ANN-stabilization parameter matches the already optimal, theoretical one for r=1, leading to nodally exact numerical solutions Instead, for higher polynomials degrees r>1, remarkable differences are observed among the theoretical and optimal stabilization parameter: we observed better accuracy and stabilization properties of the numerical solution with the optimal stabilization parameter with respect to the theoretical one.
Then, we generated the dataset on a 2D advection-diffusion problem, and we used it to train a fully-connected feed-forward ANN. We applied the predictions of the network on the original advection-diffusion problem for unseen input values and we also apply it to an unseen advection-diffusion problem to check for model generalization. Our numerical results showed that the proposed ANN-based approach provides more accurate numerical solutions than using the theoretical stabilization parameter for the SUPG method.
This work represents a step towards the enhancements of stabilization methods for the FE approximation of advection-dominated differential problems. In particular, this work is limited to few features for the ANN-based stabilization parameter and provides as output a single optimal stabilization parameter meant for the whole mesh FE Th. Natural extensions of this work will therefore involve local stabilization parameters, i.e., element by element over the mesh Th, to better account for non-uniform meshes, differential problems with varying coefficients, and capturing the local behaviour of the solution. In addition, to enhance the robustness of the ANN-based approach, possible additional inputs of the network are the angle of the advection velocity and the nodal value of the forcing term. A future development consists in the extension of the proposed approach to the 3D case for which, as for the 2D case, a universal definition of the stabilization parameter does not exist. The extension to the 3D case surely would present a larger cost in the training phase; however, once trained, the ANN could be used online real-time.
Finally, we envision extending the proposed machine learning approach to SUPG and variational multiscale stabilization methods for the FE approximation of the Navier-Stokes equations.
The authors acknowledge Prof. C. Canuto, Politecnico di Torino, for the fruitful discussions and useful suggestions. L. D. and A. Z. are members of the INdAM Research group GNCS. L. D. has been partially funded by the research project PRIN 2020, n.20204LN5N5, by MIUR. The authors ackwnowledge the anonymous reviewers for their insightful comments and suggestions.
The authors declare no conflict of interest.
[1] |
O'Brien J, Hayder H, Zayed Y, et al. (2018) Overview of microRNA biogenesis, mechanisms of actions, and circulation. Front Endocrinol 9: 402. doi: 10.3389/fendo.2018.00402
![]() |
[2] |
Cardinal-Fernández P, Ferruelo A, Esteban A, et al. (2016) Characteristics of microRNAs and their potential relevance for the diagnosis and therapy of the acute respiratory distress syndrome: from bench to bedside. Transl Res 169: 102-111. doi: 10.1016/j.trsl.2015.11.004
![]() |
[3] |
Dexheimer PJ, Cochella L (2020) MicroRNAs: from mechanism to organism. Front Cell Dev Biol 8: 409. doi: 10.3389/fcell.2020.00409
![]() |
[4] |
Nazimek K, Bryniarski K (2020) Approaches to inducing antigen-specific immune tolerance in allergy and autoimmunity: Focus on antigen-presenting cells and extracellular vesicles. Scand J Immunol 91: e12881. doi: 10.1111/sji.12881
![]() |
[5] |
Nazimek K, Bryniarski K (2020) Perspectives in manipulating EVs for therapeutic applications: focus on cancer treatment. Int J Mol Sci 21: 4623. doi: 10.3390/ijms21134623
![]() |
[6] |
Monticelli S, Ansel KM, Xiao C, et al. (2005) MicroRNA profiling of the murine hematopoietic system. Genome Biol 6: R71. doi: 10.1186/gb-2005-6-8-r71
![]() |
[7] |
Zhou B, Wang S, Mayr C, et al. (2007) miR-150, a microRNA expressed in mature B and T cells, blocks early B cell development when expressed prematurely. P Natl Acad Sci USA 104: 7080-7085. doi: 10.1073/pnas.0702409104
![]() |
[8] |
Xiao C, Calado DP, Galler G, et al. (2007) MiR-150 controls B cell differentiation by targeting the transcription factor c-Myb. Cell 131: 146-159. doi: 10.1016/j.cell.2007.07.021
![]() |
[9] |
He Y, Jiang X, Chen J (2014) The role of miR-150 in normal and malignant hematopoiesis. Oncogene 33: 3887-3893. doi: 10.1038/onc.2013.346
![]() |
[10] |
Guduric-Fuchs J, O'Connor A, Camp B, et al. (2012) Selective extracellular vesicle-mediated export of an overlapping set of microRNAs from multiple cell types. BMC Genomics 13: 357. doi: 10.1186/1471-2164-13-357
![]() |
[11] |
de Candia P, Torri A, Gorletta T, et al. (2013) Intracellular modulation, extracellular disposal and serum increase of MiR-150 mark lymphocyte activation. PLoS One 8: e75348. doi: 10.1371/journal.pone.0075348
![]() |
[12] |
de Candia P, Torri A, Pagani M, et al. (2014) Serum microRNAs as biomarkers of human lymphocyte activation in health and disease. Front Immunol 5: 43. doi: 10.3389/fimmu.2014.00043
![]() |
[13] |
Oboshi W, Hayashi K, Takeuchi H, et al. (2020) MicroRNA-150 suppresses p27Kip1 expression and promotes cell proliferation in HeLa human cervical cancer cells. Oncol Lett 20: 210. doi: 10.3892/ol.2020.12073
![]() |
[14] |
Sur D, Burz C, Sabarimurugan S, et al. (2020) Diagnostic and prognostic significance of miR-150 in colorectal cancer: a systematic review and meta-analysis. J Pers Med 10: 99. doi: 10.3390/jpm10030099
![]() |
[15] |
Plank M, Maltby S, Mattes J, et al. (2013) Targeting translational control as a novel way to treat inflammatory disease: the emerging role of microRNAs. Clin Exp Allergy 43: 981-999. doi: 10.1111/cea.12135
![]() |
[16] |
Rebane A, Akdis CA (2014) MicroRNAs in allergy and asthma. Curr Allergy Asthma Rep 14: 424. doi: 10.1007/s11882-014-0424-x
![]() |
[17] |
Weidner J, Bartel S, Kılıç A, et al. (2020) Spotlight on microRNAs in allergy and asthma. Allergy 76: 1661-1678. doi: 10.1111/all.14646
![]() |
[18] |
Rebane A (2015) microRNA and allergy. Adv Exp Med Biol 888: 331-352. doi: 10.1007/978-3-319-22671-2_17
![]() |
[19] |
Garbacki N, Di Valentin E, Huynh-Thu VA, et al. (2011) MicroRNAs profiling in murine models of acute and chronic asthma: A relationship with mRNAs targets. PLoS One 6: e16509. doi: 10.1371/journal.pone.0016509
![]() |
[20] |
Feng MJ, Shi F, Qiu C, et al. (2012) MicroRNA-181a, -146a and -146b in spleen CD4+ T lymphocytes play proinflammatory roles in a murine model of asthma. Int Immunopharmacol 13: 347-353. doi: 10.1016/j.intimp.2012.05.001
![]() |
[21] | Badalzadeh M, Mazinani M, Pourpak Z, et al. (2019) In vitro analysis of nine microRNAs in CD8+ T cells of asthmatic patients and the effects of two FDA-approved drugs. Iran J Allergy Asthma Immunol 18: 358-368. |
[22] |
Wang JW, Li K, Hellermann G, et al. (2012) MIR-150 suppresses lung inflammation in a mouse model of experimental asthma. World Allergy Organ J 5: S9. doi: 10.1097/01.WOX.0000411771.44473.bd
![]() |
[23] |
Wang JW, Li K, Hellermann G, et al. (2011) Regulating the regulators: microRNA and asthma. World Allergy Organ J 4: 94-103. doi: 10.1186/1939-4551-4-6-94
![]() |
[24] |
Zhang XY, Tang XY, Ma LJ, et al. (2017) Schisandrin B down-regulated lncRNA BCYRN1 expression of airway smooth muscle cells by improving miR-150 expression to inhibit the proliferation and migration of ASMC in asthmatic rats. Cell Proliferat 50: e12382. doi: 10.1111/cpr.12382
![]() |
[25] |
Zhang Q, Ni W, Li Y, et al. (2020) Analysis of altered miRNA profiling in the colon of a mouse model with β-lactoglobulin allergy. Allergol Immunopathol 48: 666-674. doi: 10.1016/j.aller.2020.05.007
![]() |
[26] |
Wąsik M, Nazimek K, Nowak B, et al. (2019) Delayed-type hypersensitivity underlying casein allergy is suppressed by extracellular vesicles carrying miR-150. Nutrients 11: 907. doi: 10.3390/nu11040907
![]() |
[27] |
Ho MHK, Wong WHS, Chang C (2014) Clinical spectrum of food allergies: a comprehensive review. Clin Rev Allergy Immu 46: 225-240. doi: 10.1007/s12016-012-8339-6
![]() |
[28] |
Vennegaard MT, Bonefeld CM, Hagedorn PH, et al. (2012) Allergic contact dermatitis induces upregulation of identical microRNAs in humans and mice. Contact Dermatitis 67: 298-305. doi: 10.1111/j.1600-0536.2012.02083.x
![]() |
[29] | Wolf J, Levis WR (2012) MicroRNA 150 in humans and murine contact sensitivity. J Drugs Dermatol 11: 1152. |
[30] |
Zheng Q, Zhou L, Mi QS (2012) MicroRNA miR-150 is involved in Vα14 invariant NKT cell development and function. J Immunol 188: 2118-2126. doi: 10.4049/jimmunol.1103342
![]() |
[31] |
Goubier A, Vocanson M, Macari C, et al. (2013) Invariant NKT cells suppress CD8(+) T-cell-mediated allergic contact dermatitis independently of regulatory CD4(+) T cells. J Invest Dermatol 133: 980-987. doi: 10.1038/jid.2012.404
![]() |
[32] |
Askenase PW, Bryniarski K, Paliwal V, et al. (2015) A subset of AID-dependent B-1a cells initiates hypersensitivity and pneumococcal pneumonia resistance. Ann NY Acad Sci 1362: 200-214. doi: 10.1111/nyas.12975
![]() |
[33] |
Mi QS, Xu YP, Qi RQ, et al. (2012) Lack of microRNA miR-150 reduces the capacity of epidermal Langerhans cell cross-presentation. Exp Dermatol 21: 876-877. doi: 10.1111/exd.12008
![]() |
[34] |
Gulati N, Løvendorf MB, Zibert JR, et al. (2015) Unique microRNAs appear at different times during the course of a delayed-type hypersensitivity reaction in human skin. Exp Dermatol 24: 953-957. doi: 10.1111/exd.12813
![]() |
[35] |
Ptak W, Nazimek K, Askenase PW, et al. (2015) From mysterious supernatant entity to miR-150 in antigen-specific exosomes: a history of hapten-specific T suppressor factor. Arch Immunol Ther Exp 63: 345-356. doi: 10.1007/s00005-015-0331-4
![]() |
[36] |
Bryniarski K, Ptak W, Jayakumar A, et al. (2013) Antigen-specific, antibody-coated, exosome-like nanovesicles deliver suppressor T-cell microRNA-150 to effector T cells to inhibit contact sensitivity. J Allergy Clin Immun 132: 170-181. doi: 10.1016/j.jaci.2013.04.048
![]() |
[37] |
Nazimek K, Askenase PW, Bryniarski K (2018) Antibody light chains dictate the specificity of contact hypersensitivity effector cell suppression mediated by exosomes. Int J Mol Sci 19: 2656. doi: 10.3390/ijms19092656
![]() |
[38] |
Nazimek K, Bryniarski K, Ptak W, et al. (2020) Orally administered exosomes suppress mouse delayed-type hypersensitivity by delivering miR-150 to antigen-primed macrophage APC targeted by exosome-surface anti-peptide antibody light chains. Int J Mol Sci 21: 5540. doi: 10.3390/ijms21155540
![]() |
[39] |
Nazimek K, Ptak W, Nowak B, et al. (2015) Macrophages play an essential role in antigen-specific immune suppression mediated by T CD8+ cell-derived exosomes. Immunology 146: 23-32. doi: 10.1111/imm.12466
![]() |
[40] |
Nazimek K, Bustos-Morán E, Blas-Rus N, et al. (2019) Syngeneic red blood cell-induced extracellular vesicles suppress delayed-type hypersensitivity to self-antigens in mice. Clin Exp Allergy 49: 1487-1499. doi: 10.1111/cea.13475
![]() |
[41] | Nazimek K, Nowak B, Marcinkiewicz J, et al. (2014) Enhanced generation of reactive oxygen intermediates by suppressor T cell-derived exosome-treated macrophages. Folia Med Cracov 54: 37-52. |
[42] |
Bryniarski K, Ptak W, Martin E, et al. (2015) Free extracellular miRNA functionally targets cells by transfecting exosomes from their companion cells. PLoS One 10: e0122991. doi: 10.1371/journal.pone.0122991
![]() |
[43] |
Huang XL, Zhang L, Li JP, et al. (2015) MicroRNA-150: A potential regulator in pathogens infection and autoimmune diseases. Autoimmunity 48: 503-510. doi: 10.3109/08916934.2015.1072518
![]() |
[44] |
Guan H, Peng R, Mao L, et al. (2020) Injured tubular epithelial cells activate fibroblasts to promote kidney fibrosis through miR-150-containing exosomes. Exp Cell Res 392: 112007. doi: 10.1016/j.yexcr.2020.112007
![]() |
[45] |
Luan J, Fu J, Chen C, et al. (2019) LNA-anti-miR-150 ameliorated kidney injury of lupus nephritis by inhibiting renal fibrosis and macrophage infiltration. Arthritis Res Ther 21: 276. doi: 10.1186/s13075-019-2044-2
![]() |
[46] |
Luan J, Fu J, Wang D, et al. (2020) miR-150-Based RNA interference attenuates tubulointerstitial fibrosis through the SOCS1/JAK/STAT pathway in vivo and in vitro. Mol Ther Nucleic Acids 22: 871-884. doi: 10.1016/j.omtn.2020.10.008
![]() |
[47] |
Zhou H, Hasni SA, Perez P, et al. (2013) miR-150 promotes renal fibrosis in lupus nephritis by downregulating SOCS1. J Am Soc Nephrol 24: 1073-1087. doi: 10.1681/ASN.2012080849
![]() |
[48] |
Du Z, Wu T, Liu L, et al. (2020) Extracellular vesicles-derived miR-150-5p secreted by adipose-derived mesenchymal stem cells inhibits CXCL1 expression to attenuate hepatic fibrosis. J Cell Mol Med 25: 701-715. doi: 10.1111/jcmm.16119
![]() |
[49] |
Honda N, Jinnin M, Kira-Etoh T, et al. (2013) miR-150 down-regulation contributes to the constitutive type I collagen overexpression in scleroderma dermal fibroblasts via the induction of integrin β3. Am J Pathol 182: 206-216. doi: 10.1016/j.ajpath.2012.09.023
![]() |
[50] |
Zidar N, Langner C, Jerala M, et al. (2020) Pathology of fibrosis in Crohn's disease-contribution to understanding its pathogenesis. Front Med 7: 167. doi: 10.3389/fmed.2020.00167
![]() |
[51] |
Ou H, Teng H, Qin Y, et al. (2020) Extracellular vesicles derived from microRNA-150-5p-overexpressing mesenchymal stem cells protect rat hearts against ischemia/reperfusion. Aging (Albany NY) 12: 12669-12683. doi: 10.18632/aging.102792
![]() |
[52] |
Yung S, Chan TM (2017) Molecular and immunological basis of tubulo-interstitial injury in lupus nephritis: a comprehensive review. Clin Rev Allergy Immu 52: 149-163. doi: 10.1007/s12016-016-8533-z
![]() |
[53] |
Okon LG, Werth VP (2013) Cutaneous lupus erythematosus: diagnosis and treatment. Best Pract Res Cl Rh 27: 391-404. doi: 10.1016/j.berh.2013.07.008
![]() |
[54] | Gensous N, Boizard-Moracchini A, Lazaro E, et al. (2020) Update on the cellular pathogenesis of lupus. Curr Opin Rheumatol In press. |
[55] | Chang A, Clark MR, Ko K (2020) Cellular aspects of the pathogenesis of lupus nephritis. Curr Opin Rheumatol In press. |
[56] |
Gorabi AM, Kiaie N, Aslani S, et al. (2020) Prospects for the potential of RNA interference in the treatment of autoimmune diseases: Small interfering RNAs in the spotlight. J Autoimmun 114: 102529. doi: 10.1016/j.jaut.2020.102529
![]() |
[57] |
Pauley KM, Cha S (2013) RNAi therapeutics in autoimmune disease. Pharmaceuticals (Basel) 6: 287-294. doi: 10.3390/ph6030287
![]() |
[58] |
Zhang H, Huang X, Ye L, et al. (2018) B cell-related circulating microRNAs with the potential value of biomarkers in the differential diagnosis, and distinguishment between the disease activity and lupus nephritis for systemic lupus erythematosus. Front Immunol 9: 1473. doi: 10.3389/fimmu.2018.01473
![]() |
[59] |
Steen SO, Iversen LV, Carlsen AL, et al. (2015) The circulating cell-free microRNA profile in systemic sclerosis is distinct from both healthy controls and systemic lupus erythematosus. J Rheumatol 42: 214-221. doi: 10.3899/jrheum.140502
![]() |
[60] | Nakhjavani M, Etemadi J, Pourlak T, et al. (2019) Plasma levels of miR-21, miR-150, miR-423 in patients with lupus nephritis. Iran J Kidney Dis 13: 198-206. |
[61] | Su YJ, Tsai NW, Kung CT, et al. (2018) Investigation of microRNA in mitochondrial apoptotic pathway in systemic lupus erythematosus. Biomed Res Int 2018: 9026357. |
[62] | Zhang M, Chen D, Zhang F, et al. (2020) Serum exosomal hsa-miR-135b-5p serves as a potential diagnostic biomarker in steroid-induced osteonecrosis of femoral head. Am J Transl Res 12: 2136-2154. |
[63] |
Méndez-Flores S, Furuzawa-Carballeda J, Hernández-Molina G, et al. (2019) MicroRNA expression in cutaneous lupus: a new window to understand its pathogenesis. Mediators Inflamm 2019: 5049245. doi: 10.1155/2019/5049245
![]() |
[64] |
Wang H, Peng W, Ouyang X, et al. (2012) Circulating microRNAs as candidate biomarkers in patients with systemic lupus erythematosus. Transl Res 160: 198-206. doi: 10.1016/j.trsl.2012.04.002
![]() |
[65] |
Carlsen AL, Schetter AJ, Nielsen CT, et al. (2013) Circulating microRNA expression profiles associated with systemic lupus erythematosus. Arthritis Rheum 65: 1324-1334. doi: 10.1002/art.37890
![]() |
[66] |
Solé C, Moliné T, Vidal M, et al. (2019) An exosomal urinary miRNA signature for early diagnosis of renal fibrosis in lupus nephritis. Cells 8: 773. doi: 10.3390/cells8080773
![]() |
[67] |
Abulaban KM, Fall N, Nunna R, et al. (2016) Relationship of cell-free urine MicroRNA with lupus nephritis in children. Pediatric Rheumatol 14: 4. doi: 10.1186/s12969-016-0064-x
![]() |
[68] |
Omidi F, Hosseini SA, Ahmadi A, et al. (2020) Discovering the signature of a lupus-related microRNA profile in the Gene Expression Omnibus repository. Lupus 29: 1321-1335. doi: 10.1177/0961203320944473
![]() |
[69] |
Chen JQ, Papp G, Póliska S, et al. (2017) MicroRNA expression profiles identify disease-specific alterations in systemic lupus erythematosus and primary Sjögren's syndrome. PLoS One 12: e0174585. doi: 10.1371/journal.pone.0174585
![]() |
[70] |
Ebert MS, Sharp PA (2010) MicroRNA sponges: progress and possibilities. RNA 16: 2043-2050. doi: 10.1261/rna.2414110
![]() |
[71] |
Luan J, Jiao C, Kong W, et al. (2018) circHLA-C plays an important role in lupus nephritis by sponging miR-150. Mol Ther Nucleic Acids 10: 245-253. doi: 10.1016/j.omtn.2017.12.006
![]() |
[72] |
Lu MC, Lai NS, Chen HC, et al. (2013) Decreased microRNA(miR)-145 and increased miR-224 expression in T cells from patients with systemic lupus erythematosus involved in lupus immunopathogenesis. Clin Exp Immunol 171: 91-99. doi: 10.1111/j.1365-2249.2012.04676.x
![]() |
[73] |
Voynova EN, Skinner J, Bolland S (2015) Expansion of an atypical NK cell subset in mouse models of systemic lupus erythematosus. J Immunol 194: 1503-1513. doi: 10.4049/jimmunol.1402673
![]() |
[74] |
Gao S, Yuan L, Wang Y, et al. (2017) Enhanced expression of TREM-1 in splenic cDCs in lupus prone mice and it was modulated by miR-150. Mol Immunol 81: 127-134. doi: 10.1016/j.molimm.2016.12.006
![]() |
[75] |
Fox RI (2011) Extraglandular manifestations of Sjögren's Syndrome (SS): dermatologic arthritic endocrine, pulmonary, cardiovascular, gastroenterology, renal, urology, and gynecologic manifestations. Sjögren's Syndrome New York: Springer, 285-316. doi: 10.1007/978-1-60327-957-4_17
![]() |
[76] |
Lopes AP, Hillen MR, Chouri E, et al. (2018) Circulating small non-coding RNAs reflect IFN status and B cell hyperactivity in patients with primary Sjögren's syndrome. PLoS One 13: e0193157. doi: 10.1371/journal.pone.0193157
![]() |
[77] |
Ebrahimiyan H, Gharibdoost F, Aslani S, et al. (2020) microRNAs are potentially regulating the survivin gene in PBMCs from systemic sclerosis patients. Mod Rheumatol 30: 862-869. doi: 10.1080/14397595.2019.1659545
![]() |
[78] |
Heindryckx F, Binet F, Ponticos M, et al. (2016) Endoplasmic reticulum stress enhances fibrosis through IRE1α-mediated degradation of miR-150 and XBP-1 splicing. EMBO Mol Med 8: 729-744. doi: 10.15252/emmm.201505925
![]() |
[79] |
Jinnin M (2014) Various applications of microRNAs in skin diseases. J Dermatol Sci 74: 3-8. doi: 10.1016/j.jdermsci.2014.01.004
![]() |
[80] |
Luo Y, Xiao R (2020) The epigenetic regulation of scleroderma and its clinical application. Adv Exp Med Biol 1253: 375-403. doi: 10.1007/978-981-15-3449-2_13
![]() |
[81] |
Page A, Fusil F, Cosset FL (2021) Antigen-specific tolerance approach for rheumatoid arthritis: past, present and future. Joint Bone Spine 88: 105164. doi: 10.1016/j.jbspin.2021.105164
![]() |
[82] |
Churov AV, Oleinik EK, Knip M (2015) MicroRNAs in rheumatoid arthritis: altered expression and diagnostic potential. Autoimmun Rev 14: 1029-1037. doi: 10.1016/j.autrev.2015.07.005
![]() |
[83] |
Qiu M, Mo L, Li J, et al. (2020) Effects of miR-150-5p on the growth and SOCS1 expression of rheumatoid arthritis synovial fibroblasts. Clin Rheumatol 39: 909-917. doi: 10.1007/s10067-019-04894-7
![]() |
[84] |
Zhao F, Dong J, Guo J, et al. (2020) Inhibiting role of long non-coding RNA LINC01197 in inflammation in rheumatoid arthritis through the microRNA-150/THBS2 axis. Exp Cell Res 394: 112136. doi: 10.1016/j.yexcr.2020.112136
![]() |
[85] |
Niimoto T, Nakasa T, Ishikawa M, et al. (2010) MicroRNA-146a expresses in interleukin-17 producing T cells in rheumatoid arthritis patients. BMC Musculoskelet Disord 11: 209. doi: 10.1186/1471-2474-11-209
![]() |
[86] |
Ebrahimiyan H, Rezaei N, Vojdanian M, et al. (2019) microRNA involvement in the regulation of survivin in peripheral blood mononuclear cells from rheumatoid arthritis patients. Int J Rheum Dis 22: 1107-1114. doi: 10.1111/1756-185X.13520
![]() |
[87] |
Rezaeepoor M, Pourjafar M, Tahamoli-Roudsari A, et al. (2020) Altered expression of microRNAs may predict therapeutic response in rheumatoid arthritis patients. Int Immunopharmacol 83: 106404. doi: 10.1016/j.intimp.2020.106404
![]() |
[88] |
Anaparti V, Smolik I, Meng X, et al. (2017) Whole blood microRNA expression pattern differentiates patients with rheumatoid arthritis, their seropositive first-degree relatives, and healthy unrelated control subjects. Arthritis Res Ther 19: 249. doi: 10.1186/s13075-017-1459-x
![]() |
[89] |
Chen Z, Wang H, Xia Y, et al. (2018) Therapeutic potential of mesenchymal cell-derived miR-150-5p-expressing exosomes in rheumatoid arthritis mediated by the modulation of MMP14 and VEGF. J Immunol 201: 2472-2482. doi: 10.4049/jimmunol.1800304
![]() |
[90] |
Elzorkany B, Mokbel A, Gamal SM, et al. (2021) Does smoking affect level of seropositivity in RA? A post-HOC global and inter-country analysis of COMORA cohort. Rheumatol Int 41: 699-705. doi: 10.1007/s00296-021-04791-w
![]() |
[91] |
Wasén C, Ospelt C, Camponeschi A, et al. (2020) Nicotine changes the microRNA profile to regulate the FOXO memory program of CD8+ T cells in rheumatoid arthritis. Front Immunol 11: 1474. doi: 10.3389/fimmu.2020.01474
![]() |
[92] |
Magrey MN, Haqqi T, Haseeb A (2016) Identification of plasma microRNA expression profile in radiographic axial spondyloarthritis-a pilot study. Clin Rheumatol 35: 1323-1327. doi: 10.1007/s10067-015-3123-7
![]() |
[93] |
Perez-Sanchez C, Font-Ugalde P, Ruiz-Limon P, et al. (2018) Circulating microRNAs as potential biomarkers of disease activity and structural damage in ankylosing spondylitis patients. Hum Mol Genet 27: 875-890. doi: 10.1093/hmg/ddy008
![]() |
[94] |
Lerman G, Avivi C, Mardoukh C, et al. (2011) MiRNA expression in psoriatic skin: reciprocal regulation of hsa-miR-99a and IGF-1R. PLoS One 6: e20916. doi: 10.1371/journal.pone.0020916
![]() |
[95] |
Li Y, Su J, Li F, et al. (2017) MiR-150 regulates human keratinocyte proliferation in hypoxic conditions through targeting HIF-1α and VEGFA: Implications for psoriasis treatment. PLoS One 12: e0175459. doi: 10.1371/journal.pone.0175459
![]() |
[96] |
Zhu WJ, Li P, Wang L, et al. (2020) Hypoxia-inducible factor-1: A potential pharmacological target to manage psoriasis. Int Immunopharmacol 86: 106689. doi: 10.1016/j.intimp.2020.106689
![]() |
[97] |
Torri A, Carpi D, Bulgheroni E, et al. (2017) Extracellular microRNA signature of human helper T cell subsets in health and autoimmunity. J Biol Chem 292: 2903-2915. doi: 10.1074/jbc.M116.769893
![]() |
[98] |
Okoye IS, Coomes SM, Pelly VS, et al. (2014) MicroRNA-containing T-regulatory-cell-derived exosomes suppress pathogenic T helper 1 cells. Immunity 41: 89-103. doi: 10.1016/j.immuni.2014.05.019
![]() |
[99] |
Chen WX, Ren LH, Shi RH (2014) Implication of miRNAs for inflammatory bowel disease treatment: Systematic review. World J Gastrointest Pathophysiol 5: 63-70. doi: 10.4291/wjgp.v5.i2.63
![]() |
[100] |
Bian Z, Li L, Cui J, et al. (2011) Role of miR-150-targeting c-Myb in colonic epithelial disruption during dextran sulphate sodium-induced murine experimental colitis and human ulcerative colitis. J Pathol 225: 544-553. doi: 10.1002/path.2907
![]() |
[101] |
Rodríguez-Nogales A, Algieri F, Garrido-Mesa J, et al. (2018) The administration of Escherichia coli Nissle 1917 ameliorates development of DSS-induced colitis in mice. Front Pharmacol 9: 468. doi: 10.3389/fphar.2018.00468
![]() |
[102] |
Din AU, Hassan A, Zhu Y, et al. (2020) Inhibitory effect of Bifidobacterium bifidum ATCC 29521 on colitis and its mechanism. J Nutr Biochem 79: 108353. doi: 10.1016/j.jnutbio.2020.108353
![]() |
[103] |
Rodríguez-Nogales A, Algieri F, Garrido-Mesa J, et al. (2017) Differential intestinal anti-inflammatory effects of Lactobacillus fermentum and Lactobacillus salivarius in DSS mouse colitis: impact on microRNAs expression and microbiota composition. Mol Nutr Food Res 61: 1700144. doi: 10.1002/mnfr.201700144
![]() |
[104] |
Morris NL, Hammer AM, Cannon AR, et al. (2017) Dysregulation of microRNA biogenesis in the small intestine after ethanol and burn injury. BBA-Mol Basis Dis 1863: 2645-2653. doi: 10.1016/j.bbadis.2017.03.025
![]() |
[105] | Wang S, Huang Y, Zhou C, et al. (2018) The role of autophagy and related microRNAs in inflammatory bowel disease. Gastroent Res Pract 2018: 7565076. |
[106] |
Ciccacci C, Politi C, Novelli G, et al. (2016) Advances in exploring the role of microRNAs in inflammatory bowel disease. MicroRNA 5: 5-11. doi: 10.2174/2211536605666160111124812
![]() |
[107] | Luo J, Wang Y, Lan D, et al. (2018) Differential expression of serum microRNAs in glucocorticoid-resistant patients with ulcerative colitis. Int J Clin Exp Pathol 11: 936-946. |
[108] |
Bao Y, Guo Y, Li Z, et al. (2014) MicroRNA profiling in Muc2 knockout mice of colitis-associated cancer model reveals epigenetic alterations during chronic colitis malignant transformation. PLoS One 9: e99132. doi: 10.1371/journal.pone.0099132
![]() |
[109] |
Zamvil SS, Hauser SL (2021) Antigen presentation by B cells in multiple sclerosis. N Engl J Med 384: 378-381. doi: 10.1056/NEJMcibr2032177
![]() |
[110] |
Hu Z, Cui Y, Qiao X, et al. (2018) Silencing miR-150 ameliorates experimental autoimmune encephalomyelitis. Front Neurosci 12: 465. doi: 10.3389/fnins.2018.00465
![]() |
[111] |
Fenoglio C, Cantoni C, Riz MD, et al. (2011) Expression and genetic analysis of miRNAs involved in CD4+ cell activation in patients with multiple sclerosis. Neurosci Lett 504: 9-12. doi: 10.1016/j.neulet.2011.08.021
![]() |
[112] |
Martinelli-Boneschi F, Fenoglio C, Brambilla P, et al. (2012) MicroRNA and mRNA expression profile screening in multiple sclerosis patients to unravel novel pathogenic steps and identify potential biomarkers. Neurosci Lett 508: 4-8. doi: 10.1016/j.neulet.2011.11.006
![]() |
[113] |
Jernås M, Malmeström C, Axelsson M, et al. (2013) MicroRNA regulate immune pathways in T-cells in multiple sclerosis (MS). BMC Immunol 14: 32. doi: 10.1186/1471-2172-14-32
![]() |
[114] |
Bergman P, Piket E, Khademi M, et al. (2016) Circulating miR-150 in CSF is a novel candidate biomarker for multiple sclerosis. Neurol Neuroimmunol Neuroinflamm 3: e219. doi: 10.1212/NXI.0000000000000219
![]() |
[115] |
Quintana E, Ortega FJ, Robles-Cedeño R, et al. (2017) miRNAs in cerebrospinal fluid identify patients with MS and specifically those with lipid-specific oligoclonal IgM bands. Mult Scler 23: 1716-1726. doi: 10.1177/1352458516684213
![]() |
[116] |
Bruinsma IB, van Dijk M, Bridel C, et al. (2017) Regulator of oligodendrocyte maturation, miR-219, a potential biomarker for MS. J Neuroinflammation 14: 235. doi: 10.1186/s12974-017-1006-3
![]() |
[117] |
Shakerian L, Ghorbani S, Talebi F, et al. (2018) MicroRNA-150 targets PU.1 and regulates macrophage differentiation and function in experimental autoimmune encephalomyelitis. J Neuroimmunol 323: 167-174. doi: 10.1016/j.jneuroim.2018.06.010
![]() |
[118] |
Dolati S, Aghebati-Maleki L, Ahmadi M, et al. (2018) Nanocurcumin restores aberrant miRNA expression profile in multiple sclerosis, randomized, double-blind, placebo-controlled trial. J Cell Physiol 233: 5222-5230. doi: 10.1002/jcp.26301
![]() |
[119] |
Al-Ghezi ZZ, Miranda K, Nagarkatti M, et al. (2019) Combination of cannabinoids, Δ9-tetrahydrocannabinol and cannabidiol, ameliorates experimental multiple sclerosis by suppressing neuroinflammation through regulation of miRNA-mediated signaling pathways. Front Immunol 10: 1921. doi: 10.3389/fimmu.2019.01921
![]() |
[120] |
Nadin T, Haque A, Akil M, et al. (2019) Management of the idiopathic inflammatory myopathies. Prescriber 30: 28-33. doi: 10.1002/psb.1762
![]() |
[121] |
Cotton T, Niaki OZ, Zheng B, et al. (2021) Myositis in systemic lupus erythematosus. Lupus 30: 615-619. doi: 10.1177/0961203320988587
![]() |
[122] | Ye L, Zuo Y, Yang H, et al. (2019) Specific autoantibodies and clinical phenotypes correlate with the aberrant expression of immune-related microRNAs in dermatomyositis. J Immunol Res 2019: 2927061. |
[123] |
Punga T, Le Panse R, Andersson M, et al. (2014) Circulating miRNAs in myasthenia gravis: miR-150-5p as a new potential biomarker. Ann Clin Transl Neurol 1: 49-58. doi: 10.1002/acn3.24
![]() |
[124] |
Punga AR, Andersson M, Alimohammadi M, et al. (2015) Disease specific signature of circulating miR-150-5p and miR-21-5p in myasthenia gravis patients. J Neurol Sci 356: 90-96. doi: 10.1016/j.jns.2015.06.019
![]() |
[125] |
Molin CJ, Sabre L, Weis CA, et al. (2018) Thymectomy lowers the myasthenia gravis biomarker miR-150-5p. Neurol Neuroimmunol Neuroinflamm 5: e450. doi: 10.1212/NXI.0000000000000450
![]() |
[126] |
Westerberg E, Molin CJ, Lindblad I, et al. (2017) Physical exercise in myasthenia gravis is safe and improves neuromuscular parameters and physical performance-based measures: A pilot study. Muscle Nerve 56: 207-214. doi: 10.1002/mus.25493
![]() |
[127] |
Sabre L, Maddison P, Sadalage G, et al. (2018) Circulating microRNA miR-21-5p, miR-150-5p and miR-30e-5p correlate with clinical status in late onset myasthenia gravis. J Neuroimmunol 321: 164-170. doi: 10.1016/j.jneuroim.2018.05.003
![]() |
[128] |
Punga AR, Punga T (2018) Circulating microRNAs as potential biomarkers in myasthenia gravis patients. Ann NY Acad Sci 1412: 33-40. doi: 10.1111/nyas.13510
![]() |
[129] |
Sabre L, Punga T, Punga AR (2020) Circulating miRNAs as potential biomarkers in myasthenia gravis: tools for personalized medicine. Front Immunol 11: 213. doi: 10.3389/fimmu.2020.00213
![]() |
[130] |
Sabre L, Maddison P, Wong SH, et al. (2019) miR-30e-5p as predictor of generalization in ocular myasthenia gravis. Ann Clin Transl Neurol 6: 243-251. doi: 10.1002/acn3.692
![]() |
[131] |
Fiorillo AA, Heier CR, Huang YF, et al. (2020) Estrogen receptor, inflammatory, and FOXO transcription factors regulate expression of myasthenia gravis-associated circulating microRNAs. Front Immunol 11: 151. doi: 10.3389/fimmu.2020.00151
![]() |
[132] |
Zhong H, Lu J, Jing S, et al. (2020) Low-dose rituximab lowers serum Exosomal miR-150-5p in AChR-positive refractory myasthenia gravis patients. J Neuroimmunol 348: 577383. doi: 10.1016/j.jneuroim.2020.577383
![]() |
[133] |
Cron MA, Maillard S, Truffault F, et al. (2019) Causes and consequences of miR-150-5p dysregulation in myasthenia gravis. Front Immunol 10: 539. doi: 10.3389/fimmu.2019.00539
![]() |
[134] |
Cron MA, Guillochon É, Kusner L, et al. (2020) Role of miRNAs in normal and myasthenia gravis thymus. Front Immunol 11: 1074. doi: 10.3389/fimmu.2020.01074
![]() |
[135] |
Ke Q, Kroger CJ, Clark M, et al. (2020) Evolving antibody therapies for the treatment of type 1 diabetes. Front Immunol 11: 624568. doi: 10.3389/fimmu.2020.624568
![]() |
[136] |
Estrella S, Garcia-Diaz DF, Codner E, et al. (2016) Expression of miR-22 and miR-150 in type 1 diabetes mellitus: Possible relationship with autoimmunity and clinical characteristics. Med Clin-Barcelona 147: 245-247. doi: 10.1016/j.medcli.2016.05.016
![]() |
[137] |
Wang G, Gu Y, Xu N, et al. (2018) Decreased expression of miR-150, miR146a and miR424 in type 1 diabetic patients: Association with ongoing islet autoimmunity. Biochem Biophys Res Commun 498: 382-387. doi: 10.1016/j.bbrc.2017.06.196
![]() |
[138] |
Assmann TS, Recamonde-Mendoza M, De Souza BM, et al. (2017) MicroRNA expression profiles and type 1 diabetes mellitus: systematic review and bioinformatic analysis. Endocr Connect 6: 773-790. doi: 10.1530/EC-17-0248
![]() |
[139] |
Mazzeo A, Beltramo E, Lopatina T, et al. (2018) Molecular and functional characterization of circulating extracellular vesicles from diabetic patients with and without retinopathy and healthy subjects. Exp Eye Res 176: 69-77. doi: 10.1016/j.exer.2018.07.003
![]() |
[140] |
Kim H, Bae YU, Jeon JS, et al. (2019) The circulating exosomal microRNAs related to albuminuria in patients with diabetic nephropathy. J Transl Med 17: 236. doi: 10.1186/s12967-019-1983-3
![]() |
[141] |
Lee WC, Li LC, Ng HY, et al. (2020) Urinary exosomal microRNA signatures in nephrotic, biopsy-proven diabetic nephropathy. J Clin Med 9: 1220. doi: 10.3390/jcm9041220
![]() |
[142] |
Mazzeo A, Lopatina T, Gai C, et al. (2019) Functional analysis of miR-21-3p, miR-30b-5p and miR-150-5p shuttled by extracellular vesicles from diabetic subjects reveals their association with diabetic retinopathy. Exp Eye Res 184: 56-63. doi: 10.1016/j.exer.2019.04.015
![]() |
[143] |
Henriques-Antunes H, Cardoso RMS, Zonari A, et al. (2019) The kinetics of small extracellular vesicle delivery impacts skin tissue regeneration. ACS Nano 13: 8694-8707. doi: 10.1021/acsnano.9b00376
![]() |
[144] |
Tian J, Pan W, Xu X, et al. (2020) NF-κB inhibits the occurrence of type 1 diabetes through microRNA-150-dependent PUMA degradation. Life Sci 255: 117724. doi: 10.1016/j.lfs.2020.117724
![]() |
[145] |
Roat R, Hossain MM, Christopherson J, et al. (2019) Circulating miR-150-5p is associated with immune-mediated early β-cell loss in a humanized mouse model. Xenotransplantation 26: e12474. doi: 10.1111/xen.12474
![]() |
[146] |
Hamada S, Masamune A, Kanno A, et al. (2015) Comprehensive analysis of serum microRNAs in autoimmune pancreatitis. Digestion 91: 263-271. doi: 10.1159/000381283
![]() |
[147] |
Wasik U, Kempinska-Podhorodecka A, Bogdanos DP, et al. (2020) Enhanced expression of miR-21 and miR-150 is a feature of anti-mitochondrial antibody-negative primary biliary cholangitis. Mol Med 26: 8. doi: 10.1186/s10020-019-0130-1
![]() |
[148] |
Xing L, Xu W, Qu Y, et al. (2018) miR-150 regulates B lymphocyte in autoimmune hemolytic anemia/Evans syndrome by c-Myb. Int J Hematol 107: 666-672. doi: 10.1007/s12185-018-2429-z
![]() |
[149] |
Lucherini OM, Obici L, Ferracin M, et al. (2013) First report of circulating microRNAs in tumour necrosis factor receptor-associated periodic syndrome (TRAPS). PLoS One 8: e73443. doi: 10.1371/journal.pone.0073443
![]() |
[150] | Nazimek K, Filipczak-Bryniarska I, Bryniarski K (2015) The role of medicaments, exosomes and miRNA molecules in modulation of macrophage immune activity. Postepy Hig Med Dosw 69: 1114-1129. |
[151] |
Nazimek K, Bryniarski K (2012) The biological activity of macrophages in health and disease. Postepy Hig Med Dosw 66: 507-520. doi: 10.5604/17322693.1004080
![]() |
[152] |
Trifari S, Pipkin ME, Bandukwala HS, et al. (2013) MicroRNA-directed program of cytotoxic CD8+ T-cell differentiation. P Natl Acad Sci USA 110: 18608-18613. doi: 10.1073/pnas.1317191110
![]() |
[153] |
Liang Y, Pan HF, Ye DQ (2015) microRNAs function in CD8+ T cell biology. J Leukocyte Biol 97: 487-497. doi: 10.1189/jlb.1RU0814-369R
![]() |
[154] |
Ma Z, Shen Y, Zeng Q, et al. (2018) MiR-150-5p regulates EGR2 to promote the development of chronic rhinosinusitis via the DC-Th axis. Int Immunopharmacol 54: 188-197. doi: 10.1016/j.intimp.2017.11.011
![]() |
[155] | Nazimek K, Bryniarski K, Santocki M, et al. (2015) Exosomes as mediators of intercellular communication: clinical implications. Pol Arch Med Wewn 125: 370-380. |
[156] |
Sang W, Sun C, Zhang C, et al. (2016) MicroRNA-150 negatively regulates the function of CD4(+) T cells through AKT3/Bim signaling pathway. Cell Immunol 306–307: 35-40. doi: 10.1016/j.cellimm.2016.05.007
![]() |
[157] |
Sang W, Wang Y, Zhang C, et al. (2016) MiR-150 impairs inflammatory cytokine production by targeting ARRB-2 after blocking CD28/B7 costimulatory pathway. Immunol Lett 172: 1-10. doi: 10.1016/j.imlet.2015.11.001
![]() |
[158] |
Tung SL, Boardman DA, Sen M, et al. (2018) Regulatory T cell-derived extracellular vesicles modify dendritic cell function. Sci Rep 8: 6065. doi: 10.1038/s41598-018-24531-8
![]() |
[159] |
Neamah WH, Singh NP, Alghetaa H, et al. (2019) AhR activation leads to massive mobilization of myeloid-derived suppressor cells with immunosuppressive activity through regulation of CXCR2 and microRNA miR-150-5p and miR-543-3p that target anti-inflammatory genes. J Immunol 203: 1830-1844. doi: 10.4049/jimmunol.1900291
![]() |
[160] |
Warth SC, Hoefig KP, Hiekel A, et al. (2015) Induced miR-99a expression represses Mtor cooperatively with miR-150 to promote regulatory T-cell differentiation. EMBO J 34: 1195-1213. doi: 10.15252/embj.201489589
![]() |
[161] |
Kluiver JL, Chen C-Z (2012) MicroRNAs regulate B-cell receptor signaling-induced apoptosis. Genes Immun 13: 239-244. doi: 10.1038/gene.2012.1
![]() |
[162] |
Mraz M, Chen L, Rassenti LZ, et al. (2014) miR-150 influences B-cell receptor signaling in chronic lymphocytic leukemia by regulating expression of GAB1 and FOXP1. Blood 124: 84-95. doi: 10.1182/blood-2013-09-527234
![]() |
[163] |
Cerna K, Oppelt J, Chochola V, et al. (2019) MicroRNA miR-34a downregulates FOXP1 during DNA damage response to limit BCR signalling in chronic lymphocytic leukaemia B cells. Leukemia 33: 403-414. doi: 10.1038/s41375-018-0230-x
![]() |
[164] |
Musilova K, Devan J, Cerna K, et al. (2018) miR-150 downregulation contributes to the high-grade transformation of follicular lymphoma by upregulating FOXP1 levels. Blood 132: 2389-2400. doi: 10.1182/blood-2018-06-855502
![]() |
[165] |
Jiang XX, Liu Y, Li H, et al. (2016) MYSM1/miR-150/FLT3 inhibits B1a cell proliferation. Oncotarget 7: 68086-68096. doi: 10.18632/oncotarget.11738
![]() |
[166] |
Ma Y, Liu Y, Hou H, et al. (2018) MiR-150 predicts survival in patients with sepsis and inhibits LPS-induced inflammatory factors and apoptosis by targeting NF-κB1 in human umbilical vein endothelial cells. Biochem Biophys Res Commun 500: 828-837. doi: 10.1016/j.bbrc.2018.04.168
![]() |
[167] |
Palagani A, Op de Beeck K, Naulaerts S, et al. (2014) Ectopic microRNA-150-5p transcription sensitizes glucocorticoid therapy response in MM1S multiple myeloma cells but fails to overcome hormone therapy resistance in MM1R cells. PLoS One 9: e113842. doi: 10.1371/journal.pone.0113842
![]() |
[168] |
Nazimek K, Bustos-Morán E, Blas-Rus N, et al. (2021) Antibodies enhance the suppressive activity of extracellular vesicles in mouse delayed-type hypersensitivity. Pharmaceuticals 14: 734. doi: 10.3390/ph14080734
![]() |
1. | Ignacio Brevis, Ignacio Muga, Kristoffer G. van der Zee, Neural control of discrete weak formulations: Galerkin, least squares & minimal-residual methods with quasi-optimal weights, 2022, 402, 00457825, 115716, 10.1016/j.cma.2022.115716 | |
2. | Sangeeta Yadav, Sashikumaar Ganesan, Ai-Augmented Stabilized Finite Element Method, 2023, 1556-5068, 10.2139/ssrn.4353564 | |
3. | Teeratorn Kadeethum, Daniel O’Malley, Francesco Ballarin, Ida Ang, Jan N. Fuhg, Nikolaos Bouklas, Vinicius L. S. Silva, Pablo Salinas, Claire E. Heaney, Christopher C. Pain, Sanghyun Lee, Hari S. Viswanathan, Hongkyu Yoon, Enhancing high-fidelity nonlinear solver with reduced order model, 2022, 12, 2045-2322, 10.1038/s41598-022-22407-6 | |
4. | Paola F. Antonietti, Matteo Caldana, Luca Dede’, Accelerating Algebraic Multigrid Methods via Artificial Neural Networks, 2023, 51, 2305-221X, 1, 10.1007/s10013-022-00597-w | |
5. | Axel Klawonn, Martin Lanser, Janine Weber, Learning adaptive coarse basis functions of FETI-DP, 2024, 496, 00219991, 112587, 10.1016/j.jcp.2023.112587 | |
6. | Ignacio Brevis, Ignacio Muga, David Pardo, Oscar Rodriguez, Kristoffer G. van der Zee, Learning quantities of interest from parametric PDEs: An efficient neural-weighted Minimal Residual approach, 2024, 164, 08981221, 139, 10.1016/j.camwa.2024.04.006 | |
7. | Matteo Caldana, Paola F. Antonietti, Luca Dede', Discovering artificial viscosity models for discontinuous Galerkin approximation of conservation laws using physics-informed machine learning, 2025, 520, 00219991, 113476, 10.1016/j.jcp.2024.113476 | |
8. | Léo Bois, Emmanuel Franck, Laurent Navoret, Vincent Vigon, An Optimal Control Deep Learning Method to Design Artificial Viscosities for Discontinuous Galerkin Schemes, 2024, 101, 0885-7474, 10.1007/s10915-024-02698-9 | |
9. | Tsung-Yeh Hsieh, Tsung-Hui Huang, A multiscale stabilized physics informed neural networks with weakly imposed boundary conditions transfer learning method for modeling advection dominated flow, 2024, 40, 0177-0667, 3353, 10.1007/s00366-024-01981-5 | |
10. | Sangeeta Yadav, Sashikumaar Ganesan, Artificial neural network-augmented stabilized finite element method, 2024, 499, 00219991, 112702, 10.1016/j.jcp.2023.112702 | |
11. | ||
12. | Johannes Haubner, Ottar Hellan, Marius Zeinhofer, Miroslav Kuchta, Learning mesh motion techniques with application to fluid–structure interaction, 2024, 424, 00457825, 116890, 10.1016/j.cma.2024.116890 | |
13. | ||
14. | ||
15. | ||
16. | ||
17. | ||
18. | ||
19. | ||
20. | ||
21. | ||
22. | ||
23. | ||
24. | ||
25. | ||
26. | ||
27. | ||
28. | ||
29. | ||
30. | ||
31. | ||
32. | ||
33. | ||
34. | ||
35. | ||
36. | ||
37. | ||
38. | ||
39. | ||
40. | ||
41. | ||
42. | ||
43. | ||
44. | ||
45. | ||
46. | ||
47. | ||
48. | ||
49. | ||
50. | ||
51. | ||
52. | ||
53. | ||
54. | ||
55. | ||
56. | ||
57. | ||
58. | ||
59. | ||
60. | ||
61. | ||
62. | ||
63. | ||
64. | ||
65. | ||
66. | ||
67. | ||
68. | ||
69. | ||
70. | ||
71. | ||
72. | ||
73. | ||
74. | ||
75. | ||
76. | ||
77. | ||
78. | ||
79. | ||
80. | ||
81. | ||
82. | ||
83. | ||
84. | ||
85. | ||
86. | ||
87. | ||
88. | ||
89. | ||
90. | ||
91. | ||
92. | ||
93. | ||
94. | ||
95. | ||
96. | ||
97. | ||
98. | ||
99. | ||
100. | ||
101. | ||
102. | ||
103. | ||
104. | ||
105. | ||
106. | ||
107. | ||
108. | ||
109. | ||
110. | ||
111. | ||
112. | ||
113. | ||
114. | ||
115. | ||
116. | ||
117. | ||
118. | ||
119. | ||
120. | ||
121. | ||
122. | ||
123. | ||
124. | ||
125. | ||
126. | ||
127. | ||
128. | ||
129. | ||
130. | ||
131. | ||
132. | ||
133. | ||
134. | ||
135. | ||
136. | ||
137. | ||
138. | ||
139. | ||
140. | ||
141. | ||
142. | ||
143. | ||
144. | ||
145. | ||
146. | ||
147. | ||
148. | ||
149. | ||
150. | ||
151. | ||
152. | ||
153. | ||
154. |
# Data set (m) | # Training set | # Validation set | r | h | Peg |
900 | 720(80%) | 180(20%) | {1,2,3} | {√210,√220,√240} | randomly in [7,70′710]. |
˜τr (e=˜uh−u) | τANN (e=uANNh−u) | |||||
r | h | Peh | ||e||L2(Ω) | ||e||H1(Ω) | ||e||L2(Ω) | ||e||H1(Ω) |
1 | √2/10 | 2 | 9.56⋅10−2 | 7.93⋅10−1 | 6.49⋅10−2 | 5.31⋅10−1 |
2 | √2/10 | 2 | 6.48⋅10−2 | 5.31⋅10−1 | 6.50⋅10−2 | 5.33⋅10−1 |
3 | √2/10 | 2 | 6.50⋅10−2 | 5.33⋅10−1 | 6.49⋅10−2 | 5.32⋅10−1 |
1 | √2/20 | 500 | 2.99⋅10−4 | 5.30⋅10−3 | 3.11⋅10−5 | 5.50⋅10−4 |
2 | √2/20 | 500 | 2.71⋅10−2 | 4.86⋅10−1 | 2.50⋅10−2 | 4.48⋅10−1 |
3 | √2/20 | 500 | 1.05⋅10−2 | 1.90⋅10−1 | 1.68⋅10−3 | 3.46⋅10−2 |
# Data set (m) | # Training set | # Validation set | r | h | Peg |
900 | 720(80%) | 180(20%) | {1,2,3} | {√210,√220,√240} | randomly in [7,70′710]. |
˜τr (e=˜uh−u) | τANN (e=uANNh−u) | |||||
r | h | Peh | ||e||L2(Ω) | ||e||H1(Ω) | ||e||L2(Ω) | ||e||H1(Ω) |
1 | √2/10 | 2 | 9.56⋅10−2 | 7.93⋅10−1 | 6.49⋅10−2 | 5.31⋅10−1 |
2 | √2/10 | 2 | 6.48⋅10−2 | 5.31⋅10−1 | 6.50⋅10−2 | 5.33⋅10−1 |
3 | √2/10 | 2 | 6.50⋅10−2 | 5.33⋅10−1 | 6.49⋅10−2 | 5.32⋅10−1 |
1 | √2/20 | 500 | 2.99⋅10−4 | 5.30⋅10−3 | 3.11⋅10−5 | 5.50⋅10−4 |
2 | √2/20 | 500 | 2.71⋅10−2 | 4.86⋅10−1 | 2.50⋅10−2 | 4.48⋅10−1 |
3 | √2/20 | 500 | 1.05⋅10−2 | 1.90⋅10−1 | 1.68⋅10−3 | 3.46⋅10−2 |