Chemical Property Prediction Under Experimental Biases

TL;DR

This paper addresses the challenge of bias in chemical property prediction datasets by applying causal inference techniques with graph neural networks, improving model robustness across various bias scenarios.

Contribution

It introduces the use of inverse propensity scoring and counter-factual regression methods with graph neural networks to mitigate experimental biases in chemical property prediction.

Findings

Both methods significantly improved prediction accuracy under biased conditions.

The approaches demonstrated robustness across four different bias scenarios.

Results suggest potential for more reliable chemical property modeling.

Abstract

Predicting the chemical properties of compounds is crucial in discovering novel materials and drugs with specific desired characteristics. Recent significant advances in machine learning technologies have enabled automatic predictive modeling from past experimental data reported in the literature. However, these datasets are often biased because of various reasons, such as experimental plans and publication decisions, and the prediction models trained using such biased datasets often suffer from over-fitting to the biased distributions and perform poorly on subsequent uses. Hence, this study focused on mitigating bias in the experimental datasets. We adopted two techniques from causal inference combined with graph neural networks that can represent molecular structures. The experimental results in four possible bias scenarios indicated that the inverse propensity scoring-based method and the counter-factual regression-based method made solid improvements.