B-XAIC Dataset: Benchmarking Explainable AI for Graph Neural Networks Using Chemical Data (2025-05-28T00:00:00.000000Z)

TL;DR

B-XAIC is introduced, a novel benchmark constructed from real-world molecular data and diverse tasks with known ground-truth rationales for assigned labels that reveals limitations of existing XAI methods for Graph Neural Networks (GNNs) in the molecular domain.

Abstract

Understanding the reasoning behind deep learning model predictions is crucial in cheminformatics and drug discovery, where molecular design determines their properties. However, current evaluation frameworks for Explainable AI (XAI) in this domain often rely on artificial datasets or simplified tasks, employing data-derived metrics that fail to capture the complexity of real-world scenarios and lack a direct link to explanation faithfulness. To address this, we introduce B-XAIC, a novel benchmark constructed from real-world molecular data and diverse tasks with known ground-truth rationales for assigned labels. Through a comprehensive evaluation using B-XAIC, we reveal limitations of existing XAI methods for Graph Neural Networks (GNNs) in the molecular domain. This benchmark provides a valuable resource for gaining deeper insights into the faithfulness of XAI, facilitating the development of more reliable and interpretable models.

Authors

Magdalena Proszewska

1 papers

Tomasz Danel

1 papers

Dawid Rymarczyk

1 papers

TL;DR

Abstract

Authors

References52 items

This looks like what? Challenges and Future Research Directions for Part-Prototype Models

From Black Boxes to Actionable Insights: A Perspective on Explainable Artificial Intelligence for Scientific Discovery

Towards Robust Fidelity for Evaluating Explainability of Graph Neural Networks

Explaining the Explainers in Graph Neural Networks: a Comparative Study

Global Explainability of GNNs via Logic Combination of Learned Concepts

ProGReST: Prototypical Graph Regression Soft Trees for Molecular Property Prediction

Evaluating explainability for graph neural networks

GREASE: Generate Factual and Counterfactual Explanations for GNN-based Recommendations

GraphFramEx: Towards Systematic Evaluation of Explainability Methods for Graph Neural Networks

Learning and Evaluating Graph Neural Network Explanations based on Counterfactual and Factual Reasoning

Interpretable and Generalizable Graph Learning via Stochastic Attention Mechanism

Discovering Invariant Rationales for Graph Neural Networks

From Anecdotal Evidence to Quantitative Evaluation Methods: A Systematic Review on Evaluating Explainable AI

KerGNNs: Interpretable Graph Neural Networks with Graph Kernels

Improving Subgraph Recognition with Variational Graph Information Bottleneck

Graph Kernel Neural Networks

ProtGNN: Towards Self-Explaining Graph Neural Networks

Towards Self-Explainable Graph Neural Network

When Comparing to Ground Truth is Wrong: On Evaluating GNN Explanation Methods

Robust Counterfactual Explanations on Graph Neural Networks

Towards a Unified Framework for Fair and Stable Graph Representation Learning

On Explainability of Graph Neural Networks via Subgraph Explorations

CF-GNNExplainer: Counterfactual Explanations for Graph Neural Networks

Explainability in Graph Neural Networks: A Taxonomic Survey

A compact review of molecular property prediction with graph neural networks.

Parameterized Explainer for Graph Neural Network

Graph Information Bottleneck

Interpreting Graph Neural Networks for NLP With Differentiable Edge Masking

Drug discovery with explainable artificial intelligence

RelEx: A Model-Agnostic Relational Model Explainer

Understanding Attention and Generalization in Graph Neural Networks

GNNExplainer: Generating Explanations for Graph Neural Networks

Using attribution to decode binding mechanism in neural network models for chemistry

Graph Attention Networks

Axiomatic Attribution for Deep Networks

Salient Deconvolutional Networks

Semi-Supervised Classification with Graph Convolutional Networks

Not Just a Black Box: Learning Important Features Through Propagating Activation Differences

DeepTox: Toxicity Prediction using Deep Learning

Striving for Simplicity: The All Convolutional Net

Deep Inside Convolutional Networks: Visualising Image Classification Models and Saliency Maps

A Bayesian Approach to in Silico Blood-Brain Barrier Penetration Modeling

ChEMBL: a large-scale bioactivity database for drug discovery

An Efficient Explanation of Individual Classifications using Game Theory

New substructure filters for removal of pan assay interference compounds (PAINS) from screening libraries and for their exclusion in bioassays.

Comparison of descriptor spaces for chemical compound retrieval and classification

Structure-activity relationship of mutagenic aromatic and heteroaromatic nitro compounds. Correlation with molecular orbital energies and hydrophobicity.

Figures

Evaluating Attribution for Graph Neural Networks

7th International Conference on Learning Representations, ICLR 2019, New Orleans, LA, USA, May 6-9, 2019

Protein function prediction via graph kernels

Detection of indoles, a bicyclic structure with a benzene ring fused to a pyrrole ring

Field of Study

Journal Information

Name

Volume

Venue Information

Name

Type

URL

Alternate Names