1
|
Mastropietro A, Bajorath J. Protocol to explain support vector machine predictions via exact Shapley value computation. STAR Protoc 2024; 5:103010. [PMID: 38607924 PMCID: PMC11017346 DOI: 10.1016/j.xpro.2024.103010] [Citation(s) in RCA: 0] [Impact Index Per Article: 0] [Reference Citation Analysis] [What about the content of this article? (0)] [Affiliation(s)] [Abstract] [Key Words] [Track Full Text] [Journal Information] [Subscribe] [Scholar Register] [Received: 02/19/2024] [Revised: 03/11/2024] [Accepted: 03/25/2024] [Indexed: 04/14/2024] Open
Abstract
Shapley values from cooperative game theory are adapted for explaining machine learning predictions. For large feature sets used in machine learning, Shapley values are approximated. We present a protocol for two techniques for explaining support vector machine predictions with exact Shapley value computation. We detail the application of these algorithms and provide ready-to-use Python scripts and custom code. The final output of the protocol includes quantitative feature analysis and mapping of important features for visualization. For complete details on the use and execution of this protocol, please refer to Feldmann and Bajorath1 and Mastropietro et al.2.
Collapse
Affiliation(s)
- Andrea Mastropietro
- Deparment of Computer, Control and Management Engineering "Antonio Ruberti", Sapienza University of Rome, Via Ariosto 25, 00185 Rome, Italy.
| | - Jürgen Bajorath
- Department of Life Science Informatics and Data Science, B-IT, LIMES Program Unit Chemical Biology and Medicinal Chemistry, Rheinische Friedrich-Wilhelms-Universität, Friedrich-Hirzebruch-Allee 5/6, 53115 Bonn, Germany; Lamarr Institute for Machine Learning and Artificial Intelligence, Friedrich-Hirzebruch-Allee 5/6, 53115 Bonn, Germany.
| |
Collapse
|
2
|
Pirovano I, Antonacci Y, Mastropietro A, Bara C, Sparacino L, Guanziroli E, Molteni F, Tettamanti M, Faes L, Rizzo G. Rehabilitation Modulates High-Order Interactions Among Large-Scale Brain Networks in Subacute Stroke. IEEE Trans Neural Syst Rehabil Eng 2023; 31:4549-4560. [PMID: 37955999 DOI: 10.1109/tnsre.2023.3332114] [Citation(s) in RCA: 0] [Impact Index Per Article: 0] [Reference Citation Analysis] [What about the content of this article? (0)] [Abstract] [MESH Headings] [Track Full Text] [Journal Information] [Subscribe] [Scholar Register] [Indexed: 11/15/2023]
Abstract
The recovery of motor functions after stroke is fostered by the functional integration of large-scale brain networks, including the motor network (MN) and high-order cognitive controls networks, such as the default mode (DMN) and executive control (ECN) networks. In this paper, electroencephalography signals are used to investigate interactions among these three resting state networks (RSNs) in subacute stroke patients after motor rehabilitation. A novel metric, the O-information rate (OIR), is used to quantify the balance between redundancy and synergy in the complex high-order interactions among RSNs, as well as its causal decomposition to identify the direction of information flow. The paper also employs conditional spectral Granger causality to assess pairwise directed functional connectivity between RSNs. After rehabilitation, a synergy increase among these RSNs is found, especially driven by MN. From the pairwise description, a reduced directed functional connectivity towards MN is enhanced after treatment. Besides, inter-network connectivity changes are associated with motor recovery, for which the mediation role of ECN seems to play a relevant role, both from pairwise and high-order interactions perspective.
Collapse
|
3
|
Mastropietro A, Feldmann C, Bajorath J. Calculation of exact Shapley values for explaining support vector machine models using the radial basis function kernel. Sci Rep 2023; 13:19561. [PMID: 37949930 PMCID: PMC10638308 DOI: 10.1038/s41598-023-46930-2] [Citation(s) in RCA: 2] [Impact Index Per Article: 2.0] [Reference Citation Analysis] [What about the content of this article? (0)] [Affiliation(s)] [Abstract] [Key Words] [Grants] [Track Full Text] [Journal Information] [Subscribe] [Scholar Register] [Received: 06/01/2023] [Accepted: 11/07/2023] [Indexed: 11/12/2023] Open
Abstract
Machine learning (ML) algorithms are extensively used in pharmaceutical research. Most ML models have black-box character, thus preventing the interpretation of predictions. However, rationalizing model decisions is of critical importance if predictions should aid in experimental design. Accordingly, in interdisciplinary research, there is growing interest in explaining ML models. Methods devised for this purpose are a part of the explainable artificial intelligence (XAI) spectrum of approaches. In XAI, the Shapley value concept originating from cooperative game theory has become popular for identifying features determining predictions. The Shapley value concept has been adapted as a model-agnostic approach for explaining predictions. Since the computational time required for Shapley value calculations scales exponentially with the number of features used, local approximations such as Shapley additive explanations (SHAP) are usually required in ML. The support vector machine (SVM) algorithm is one of the most popular ML methods in pharmaceutical research and beyond. SVM models are often explained using SHAP. However, there is only limited correlation between SHAP and exact Shapley values, as previously demonstrated for SVM calculations using the Tanimoto kernel, which limits SVM model explanation. Since the Tanimoto kernel is a special kernel function mostly applied for assessing chemical similarity, we have developed the Shapley value-expressed radial basis function (SVERAD), a computationally efficient approach for the calculation of exact Shapley values for SVM models based upon radial basis function kernels that are widely applied in different areas. SVERAD is shown to produce meaningful explanations of SVM predictions.
Collapse
Affiliation(s)
- Andrea Mastropietro
- Department of Computer, Control and Management Engineering "Antonio Ruberti", Sapienza University of Rome, 00185, Rome, Italy
| | - Christian Feldmann
- Department of Life Science Informatics and Data Science, B-IT, LIMES Program Unit Chemical Biology and Medicinal Chemistry, Rheinische Friedrich-Wilhelms-Universität, Friedrich-Hirzebruch-Allee 5/6, 53115, Bonn, Germany
| | - Jürgen Bajorath
- Department of Life Science Informatics and Data Science, B-IT, LIMES Program Unit Chemical Biology and Medicinal Chemistry, Rheinische Friedrich-Wilhelms-Universität, Friedrich-Hirzebruch-Allee 5/6, 53115, Bonn, Germany.
| |
Collapse
|
4
|
Mastropietro A, De Carlo G, Anagnostopoulos A. XGDAG: explainable gene-disease associations via graph neural networks. Bioinformatics 2023; 39:btad482. [PMID: 37531293 PMCID: PMC10421968 DOI: 10.1093/bioinformatics/btad482] [Citation(s) in RCA: 0] [Impact Index Per Article: 0] [Reference Citation Analysis] [What about the content of this article? (0)] [Affiliation(s)] [Abstract] [MESH Headings] [Grants] [Track Full Text] [Journal Information] [Subscribe] [Scholar Register] [Received: 04/11/2023] [Revised: 06/27/2023] [Accepted: 08/01/2023] [Indexed: 08/04/2023] Open
Abstract
MOTIVATION Disease gene prioritization consists in identifying genes that are likely to be involved in the mechanisms of a given disease, providing a ranking of such genes. Recently, the research community has used computational methods to uncover unknown gene-disease associations; these methods range from combinatorial to machine learning-based approaches. In particular, during the last years, approaches based on deep learning have provided superior results compared to more traditional ones. Yet, the problem with these is their inherent black-box structure, which prevents interpretability. RESULTS We propose a new methodology for disease gene discovery, which leverages graph-structured data using graph neural networks (GNNs) along with an explainability phase for determining the ranking of candidate genes and understanding the model's output. Our approach is based on a positive-unlabeled learning strategy, which outperforms existing gene discovery methods by exploiting GNNs in a non-black-box fashion. Our methodology is effective even in scenarios where a large number of associated genes need to be retrieved, in which gene prioritization methods often tend to lose their reliability. AVAILABILITY AND IMPLEMENTATION The source code of XGDAG is available on GitHub at: https://github.com/GiDeCarlo/XGDAG. The data underlying this article are available at: https://www.disgenet.org/, https://thebiogrid.org/, https://doi.org/10.1371/journal.pcbi.1004120.s003, and https://doi.org/10.1371/journal.pcbi.1004120.s004.
Collapse
Affiliation(s)
- Andrea Mastropietro
- Department of Computer, Control and Management Engineering “Antonio Ruberti”, Sapienza University of Rome, Rome 00185, Italy
| | - Gianluca De Carlo
- Department of Computer, Control and Management Engineering “Antonio Ruberti”, Sapienza University of Rome, Rome 00185, Italy
| | - Aris Anagnostopoulos
- Department of Computer, Control and Management Engineering “Antonio Ruberti”, Sapienza University of Rome, Rome 00185, Italy
| |
Collapse
|
5
|
Stolfi P, Mastropietro A, Pasculli G, Tieri P, Vergni D. NIAPU: network-informed adaptive positive-unlabeled learning for disease gene identification. Bioinformatics 2023; 39:7023926. [PMID: 36727493 PMCID: PMC9933847 DOI: 10.1093/bioinformatics/btac848] [Citation(s) in RCA: 0] [Impact Index Per Article: 0] [Reference Citation Analysis] [What about the content of this article? (0)] [Affiliation(s)] [Abstract] [Track Full Text] [Figures] [Journal Information] [Subscribe] [Scholar Register] [Received: 06/01/2022] [Revised: 12/23/2022] [Indexed: 02/03/2023] Open
Abstract
MOTIVATION Gene-disease associations are fundamental for understanding disease etiology and developing effective interventions and treatments. Identifying genes not yet associated with a disease due to a lack of studies is a challenging task in which prioritization based on prior knowledge is an important element. The computational search for new candidate disease genes may be eased by positive-unlabeled learning, the machine learning (ML) setting in which only a subset of instances are labeled as positive while the rest of the dataset is unlabeled. In this work, we propose a set of effective network-based features to be used in a novel Markov diffusion-based multi-class labeling strategy for putative disease gene discovery. RESULTS The performances of the new labeling algorithm and the effectiveness of the proposed features have been tested on 10 different disease datasets using three ML algorithms. The new features have been compared against classical topological and functional/ontological features and a set of network- and biological-derived features already used in gene discovery tasks. The predictive power of the integrated methodology in searching for new disease genes has been found to be competitive against state-of-the-art algorithms. AVAILABILITY AND IMPLEMENTATION The source code of NIAPU can be accessed at https://github.com/AndMastro/NIAPU. The source data used in this study are available online on the respective websites. SUPPLEMENTARY INFORMATION Supplementary data are available at Bioinformatics online.
Collapse
Affiliation(s)
- Paola Stolfi
- Institute for Applied Computing (IAC) 'Mauro Picone', National Research Council of Italy (CNR), Rome 00185, Italy
| | - Andrea Mastropietro
- Department of Computer, Control and Management Engineering (DIAG) 'Antonio Ruberti', Sapienza University of Rome, Rome 00185, Italy
| | - Giuseppe Pasculli
- Department of Computer, Control and Management Engineering (DIAG) 'Antonio Ruberti', Sapienza University of Rome, Rome 00185, Italy
| | - Paolo Tieri
- Institute for Applied Computing (IAC) 'Mauro Picone', National Research Council of Italy (CNR), Rome 00185, Italy
| | - Davide Vergni
- Institute for Applied Computing (IAC) 'Mauro Picone', National Research Council of Italy (CNR), Rome 00185, Italy
| |
Collapse
|
6
|
Mastropietro A, Pasculli G, Bajorath J. Protocol to explain graph neural network predictions using an edge-centric Shapley value-based approach. STAR Protoc 2022; 3:101887. [PMID: 36595907 PMCID: PMC9700376 DOI: 10.1016/j.xpro.2022.101887] [Citation(s) in RCA: 0] [Impact Index Per Article: 0] [Reference Citation Analysis] [What about the content of this article? (0)] [Affiliation(s)] [Abstract] [Key Words] [Track Full Text] [Download PDF] [Figures] [Journal Information] [Subscribe] [Scholar Register] [Received: 09/20/2022] [Revised: 10/19/2022] [Accepted: 11/04/2022] [Indexed: 11/25/2022] Open
Abstract
Here we present EdgeSHAPer, a workflow for explaining graph neural networks by approximating Shapley values using Monte Carlo sampling. In this protocol, we describe steps to execute Python scripts for a chemical dataset from the original publication; however, this approach is also applicable to any user-provided dataset. We also detail steps encompassing neural network training, an explanation phase, and analysis via feature mapping. For complete details on the use and execution of this protocol, please refer to Mastropietro et al. (2022).1.
Collapse
Affiliation(s)
- Andrea Mastropietro
- Department of Computer, Control and Management Engineering Antonio Ruberti (DIAG) Sapienza University of Rome, 00185 Rome, Italy,Department of Life Science Informatics and Data Science, B-IT, LIMES Program Unit Chemical Biology and Medicinal Chemistry, Rheinische Friedrich-Wilhelms-Universität, Friedrich-Hirzebruch-Allee 5/6, 53115 Bonn, Germany,Corresponding author
| | - Giuseppe Pasculli
- Department of Computer, Control and Management Engineering Antonio Ruberti (DIAG) Sapienza University of Rome, 00185 Rome, Italy
| | - Jürgen Bajorath
- Department of Life Science Informatics and Data Science, B-IT, LIMES Program Unit Chemical Biology and Medicinal Chemistry, Rheinische Friedrich-Wilhelms-Universität, Friedrich-Hirzebruch-Allee 5/6, 53115 Bonn, Germany,Corresponding author
| |
Collapse
|
7
|
Mastropietro A, Pasculli G, Feldmann C, Rodríguez-Pérez R, Bajorath J. EdgeSHAPer: Bond-Centric Shapley Value-Based Explanation Method for Graph Neural Networks. iScience 2022; 25:105043. [PMID: 36134335 PMCID: PMC9483788 DOI: 10.1016/j.isci.2022.105043] [Citation(s) in RCA: 8] [Impact Index Per Article: 4.0] [Reference Citation Analysis] [What about the content of this article? (0)] [Abstract] [Track Full Text] [Download PDF] [Figures] [Journal Information] [Subscribe] [Scholar Register] [Received: 06/21/2022] [Revised: 08/17/2022] [Accepted: 08/25/2022] [Indexed: 11/29/2022] Open
Abstract
Graph neural networks (GNNs) recursively propagate signals along the edges of an input graph, integrate node feature information with graph structure, and learn object representations. Like other deep neural network models, GNNs have notorious black box character. For GNNs, only few approaches are available to rationalize model decisions. We introduce EdgeSHAPer, a generally applicable method for explaining GNN-based models. The approach is devised to assess edge importance for predictions. Therefore, EdgeSHAPer makes use of the Shapley value concept from game theory. For proof-of-concept, EdgeSHAPer is applied to compound activity prediction, a central task in drug discovery. EdgeSHAPer’s edge centricity is relevant for molecular graphs where edges represent chemical bonds. Combined with feature mapping, EdgeSHAPer produces intuitive explanations for compound activity predictions. Compared to a popular node-centric and another edge-centric GNN explanation method, EdgeSHAPer reveals higher resolution in differentiating features determining predictions and identifies minimal pertinent positive feature sets. EdgeSHAPer is new methodology for explaining graph neural network models Edge centricity represents a characteristic feature of the approach EdgeSHAPer is generally applicable including molecular predictions EdgeSHAPer produces explanations of compound predictions at a high resolution
Collapse
|
8
|
Shahini E, Pasculli G, Mastropietro A, Stolfi P, Tieri P, Vergni D, Cozzolongo R, Pesce F, Giannelli G. Network Proximity-Based Drug Repurposing Strategy for Early and Late Stages of Primary Biliary Cholangitis. Biomedicines 2022; 10:1694. [PMID: 35884999 PMCID: PMC9312896 DOI: 10.3390/biomedicines10071694] [Citation(s) in RCA: 0] [Impact Index Per Article: 0] [Reference Citation Analysis] [What about the content of this article? (0)] [Affiliation(s)] [Abstract] [Key Words] [Track Full Text] [Download PDF] [Figures] [Journal Information] [Subscribe] [Scholar Register] [Received: 04/23/2022] [Revised: 06/03/2022] [Accepted: 07/11/2022] [Indexed: 11/30/2022] Open
Abstract
Primary biliary cholangitis (PBC) is a chronic, cholestatic, immune-mediated, and progressive liver disorder. Treatment to preventing the disease from advancing into later and irreversible stages is still an unmet clinical need. Accordingly, we set up a drug repurposing framework to find potential therapeutic agents targeting relevant pathways derived from an expanded pool of genes involved in different stages of PBC. Starting with updated human protein-protein interaction data and genes specifically involved in the early and late stages of PBC, a network medicine approach was used to provide a PBC "proximity" or "involvement" gene ranking using network diffusion algorithms and machine learning models. The top genes in the proximity ranking, when combined with the original PBC-related genes, resulted in a final dataset of the genes most involved in PBC disease. Finally, a drug repurposing strategy was implemented by mining and utilizing dedicated drug-gene interaction and druggable genome information knowledge bases (e.g., the DrugBank repository). We identified several potential drug candidates interacting with PBC pathways after performing an over-representation analysis on our initial 1121-seed gene list and the resulting disease-associated (algorithm-obtained) genes. The mechanism and potential therapeutic applications of such drugs were then thoroughly discussed, with a particular emphasis on different stages of PBC disease. We found that interleukin/EGFR/TNF-alpha inhibitors, branched-chain amino acids, geldanamycin, tauroursodeoxycholic acid, genistein, antioestrogens, curcumin, antineovascularisation agents, enzyme/protease inhibitors, and antirheumatic agents are promising drugs targeting distinct stages of PBC. We developed robust and transparent selection mechanisms for prioritizing already approved medicinal products or investigational products for repurposing based on recognized unmet medical needs in PBC, as well as solid preliminary data to achieve this goal.
Collapse
Affiliation(s)
- Endrit Shahini
- National Institute of Research IRCCS “Saverio De Bellis”, Castellana Grotte, 70013 Bari, Italy; (R.C.); (G.G.)
| | - Giuseppe Pasculli
- Department of Computer, Control and Management Engineering Antonio Ruberti (DIAG), Sapienza University of Rome, 00185 Rome, Italy; (G.P.); (A.M.)
| | - Andrea Mastropietro
- Department of Computer, Control and Management Engineering Antonio Ruberti (DIAG), Sapienza University of Rome, 00185 Rome, Italy; (G.P.); (A.M.)
| | - Paola Stolfi
- National Research Council (CNR), Institute for Applied Computing (IAC), 00185 Rome, Italy; (P.S.); (P.T.); (D.V.)
| | - Paolo Tieri
- National Research Council (CNR), Institute for Applied Computing (IAC), 00185 Rome, Italy; (P.S.); (P.T.); (D.V.)
| | - Davide Vergni
- National Research Council (CNR), Institute for Applied Computing (IAC), 00185 Rome, Italy; (P.S.); (P.T.); (D.V.)
| | - Raffaele Cozzolongo
- National Institute of Research IRCCS “Saverio De Bellis”, Castellana Grotte, 70013 Bari, Italy; (R.C.); (G.G.)
| | - Francesco Pesce
- Department of Emergency and Organ Transplantation, Nephrology, Dialysis and Transplantation Unit, University of Bari “A. Moro”, 70121 Bari, Italy;
| | - Gianluigi Giannelli
- National Institute of Research IRCCS “Saverio De Bellis”, Castellana Grotte, 70013 Bari, Italy; (R.C.); (G.G.)
| |
Collapse
|
9
|
Pirovano I, Mastropietro A, Guanziroli E, Molteni F, Faes L, Rizzo G. Comparison between directed causal flow metrics for the assessment of resting-state EEG motor network connectivity in subacute stroke patients. Annu Int Conf IEEE Eng Med Biol Soc 2022; 2022:44-47. [PMID: 36085760 DOI: 10.1109/embc48229.2022.9870885] [Citation(s) in RCA: 0] [Impact Index Per Article: 0] [Reference Citation Analysis] [What about the content of this article? (0)] [Abstract] [MESH Headings] [Track Full Text] [Subscribe] [Scholar Register] [Indexed: 06/15/2023]
Abstract
Isolated effective coherence (iCoh) is a measure of neural causal functional connectivity from EEG signals that was proven to overperform the Generalized Partial Directed Coherence (gPDC). However, iCoh sensitivity in the identification of reliable functional neural connections with respect to random links was not investigated. This study aims to compare the sensitivity of iCoh and gPDC with a statistical surrogates' approach. The cerebral motor network topology of a cohort of subjects in sub-acute stage after stroke was investigated. iCoh showed enhanced statistical discriminative power of the relevant connections within the motor network with respect to gPDC. This property influenced the assessment of ipsilesional intra-hemispheric topographic variations occurring in the population after a physical rehabilitation program.
Collapse
|
10
|
Scalco E, Mastropietro A, Bodini A, Marzi S, Rizzo G. A Multi-Variate framework to assess reliability and discrimination power of Bayesian estimation of Intravoxel Incoherent Motion parameters. Phys Med 2021; 89:11-19. [PMID: 34343762 DOI: 10.1016/j.ejmp.2021.07.025] [Citation(s) in RCA: 3] [Impact Index Per Article: 1.0] [Reference Citation Analysis] [What about the content of this article? (0)] [Affiliation(s)] [Abstract] [Key Words] [Track Full Text] [Journal Information] [Submit a Manuscript] [Subscribe] [Scholar Register] [Received: 03/17/2021] [Revised: 06/28/2021] [Accepted: 07/19/2021] [Indexed: 02/07/2023] Open
Abstract
PURPOSE To propose a multivariate multi-step framework for a systematic assessment of the estimation reliability and discriminability of Intravoxel Incoherent Motion (IVIM) model parameters. METHODS Monte-Carlo simulations were generated on a range of SNRs and in different IVIM combinations considering: i) a dense discretization with 24 b-values; ii) a discretization with 9 b-values. A state-of-the-art Bayesian fitting method was adopted. The framework assessed: i) the best model between mono- and bi-exponential, through the BIC index; ii) the fitting accuracy; iii) the power in discriminating two different IVIM parameters distributions of estimated coefficients, using a multivariate test. Exemplificative oncologic cases were also presented. RESULTS The bi-exponential fitting was reliable for perfusion fraction higher than 5%, with high accuracy in D estimation, acceptable error for f, but high uncertainty in D*. The discrimination of two distributions is generally feasible if differences in D values (at least 0.3 x10-3 mm2/s) are present; in the case of similar D values, a minimal difference of 5% in f can be discriminated just in case of balanced sample size and dense b-values discretization, whereas the impact of D* is quite negligible. These results were also supported by clinical examples. CONCLUSIONS IVIM model is generally accurate in estimating diffusion, but uncertainties related to perfusion estimation are not negligible and compromise the discrimination power when different populations should be differentiated. The proposed framework should be adopted as interpretative guidelines to better understand when IVIM model applied on real data can provide reliable findings.
Collapse
Affiliation(s)
- E Scalco
- Institute of Biomedical Technologies, Italian National Research Council (ITB-CNR), Segrate, Italy
| | - A Mastropietro
- Institute of Biomedical Technologies, Italian National Research Council (ITB-CNR), Segrate, Italy.
| | - A Bodini
- Institute for Applied Mathematics and Information Technologies "E. Magenes", Italian National Research Council (IMATI-CNR), Milano, Italy
| | - S Marzi
- Medical Physics Laboratory, Regina Elena National Cancer Institute, Roma, Italy
| | - G Rizzo
- Institute of Biomedical Technologies, Italian National Research Council (ITB-CNR), Segrate, Italy
| |
Collapse
|
11
|
Moroni R, Zucca I, Inverardi F, Mastropietro A, Regondi M, Spreafico R, Frassoni C. In vivo detection of cortical abnormalities in BCNU-treated rats, model of cortical dysplasia, using manganese-enhanced magnetic resonance imaging. Neuroscience 2011; 192:564-71. [DOI: 10.1016/j.neuroscience.2011.07.009] [Citation(s) in RCA: 4] [Impact Index Per Article: 0.3] [Reference Citation Analysis] [What about the content of this article? (0)] [Track Full Text] [Journal Information] [Subscribe] [Scholar Register] [Received: 05/04/2011] [Revised: 07/01/2011] [Accepted: 07/06/2011] [Indexed: 10/18/2022]
|
12
|
Garbelli R, Zucca I, Milesi G, Mastropietro A, D'Incerti L, Tassi L, Colombo N, Marras C, Villani F, Minati L, Spreafico R. Combined 7-T MRI and histopathologic study of normal and dysplastic samples from patients with TLE. Neurology 2011; 76:1177-85. [DOI: 10.1212/wnl.0b013e318212aae1] [Citation(s) in RCA: 36] [Impact Index Per Article: 2.8] [Reference Citation Analysis] [What about the content of this article? (0)] [Track Full Text] [Journal Information] [Subscribe] [Scholar Register] [Indexed: 11/15/2022] Open
|