Prediction of Clinical Features Using Graph Neural Networks

Discovery of functional motifs by association to clinical features using Graph Neural Networks.

Running

Training PNA For Regression

python train_test_controller.py --aggregators 'max' --bs 16 --dropout 0.0 --en my_experiment --epoch 200 --factor 0.8 --fcl 256 --gcn_h 64 --lr 0.001 --min_lr 0.0001 --model PNAConv --num_of_ff_layers 1 --num_of_gcn_layers 2 --patience 5 --scalers 'identity' --weight_decay 1e-05

Training GAT For Regression

python train_test_controller.py --aggregators None --bs 16 --dropout 0.0 --en my_experiment --epoch 200 --factor 0.2 --fcl 128 --gcn_h 64 --lr 0.001 --min_lr 2e-05 --model GATConv --num_of_ff_layers 1 --num_of_gcn_layers 3 --patience 20 --scalers None --weight_decay 0

GNNExplainer For PNA Regressor

python gnnexplainer.py --aggregators 'max' --bs 16 --dropout 0.0 --fcl 256 --gcn_h 64 --model PNAConv --num_of_ff_layers 1 --num_of_gcn_layers 2 --scalers 'identity' --idx 10

LIME Explainer for PNA Regressor

python lime.py --aggregators 'max' --bs 16 --dropout 0.0 --fcl 256 --gcn_h 64 --model PNAConv --num_of_ff_layers 1 --num_of_gcn_layers 2 --scalers 'identity' --idx 10

SHAP Explainer for PNA Regressor

python shap.py --aggregators 'max' --bs 16 --dropout 0.0 --fcl 256 --gcn_h 64 --model PNAConv --num_of_ff_layers 1 --num_of_gcn_layers 2 --scalers 'identity' --idx 10

Explainable cells and cell interactions

Original Graph	SubGraph

First Results of Hyperparameter Tuning

Resources

• Awesome graph explainability papers
• Towards Explainable Graph Neural Networks
• Parameterized Explainer for Graph Neural Networks (Github-PyTorch)
• Parameterized Explainer for Graph Neural Networks (Github-TensorFlow)
• GNNExplainer - DGL
• GNNExplainer on MUTAG (Graph Classification) - Colab
• GNNExplainer on MUTAG (Graph Classification) - Colab 2
• Captum - Paper
• Captum - Website
• Explainability in Graph Neural Networks
• Distributed Computation of Attributions using Captum
• Workshop on GNNExplainer with Graph Classification (MUTAG)
• Extending Gnnexplainer for graph classification - PyG
• Explainability Techniques for Graph Convolutional Networks [Code]

---

• SHAP-Library
• GraphSVX: Shapley Value Explanations for Graph Neural Networks
• The Shapley Value in Machine Learning
• GRAPHSHAP: Motif-based Explanations for Black-box Graph Classifiers
• GraphLIME: Local Interpretable Model Explanations for Graph Neural Networks
• CF-GNNExplainer: Counterfactual Explanations for Graph Neural Networks
• Explaining Graph Neural Networks with Structure-Aware Cooperative Games
• Reliable Graph Neural Network Explanations Through Adversarial Training
• Explainable AI Video Series
• awesome-machine-learning-interpretability
• PGM-Explainer: Probabilistic Graphical Model Explanations for Graph Neural Networks
• An Explainable AI Library for Benchmarking Graph Explainers-GXAI [Code]
• On Explainability of Graph Neural Networks via Subgraph Explorations [Code]
• Data Representing Ground-Truth Explanations to Evaluate XAI Methods
• Towards Ground Truth Explainability on Tabular Data
• Interpretable Machine Learning
• “Why Should I Trust You?” Explaining the Predictions of Any Classifier
• Quantus
• Papers and code of Explainable AI esp. w.r.t. Image classificiation
• OpenXAI
• How can I choose an explainer? An Application-grounded Evaluation of Post-hoc Explanations

Evaluating the Explainers

• Evaluating Attribution for Graph Neural Networks

Biological/Biomedicine Papers & Repos That used/cited GNNExplainer Paper

• histocartography
• Towards Explainable Graph Representations in Digital Pathology
• An Causal XAI Diagnostic Model for Breast Cancer Based on Mammography Reports
• Predicting Cell Type and Extracting Key Genes using Single Cell Multi-Omics Data and Graph Neural Networks
• Graph Representation Learning in Biomedicine
• Predicting the Survival of Cancer Patients With Multimodal Graph Neural Network
• scDeepSort: a pre-trained cell-type annotation method for single-cell transcriptomics using deep learning with a weighted graph neural network [Code]
• A survey on graph-based deep learning for computational histopathology
• Explaining decisions of graph convolutional neural networks: patient-specific molecular subnetworks responsible for metastasis prediction in breast cancer