MoleculeQA: A Dataset to Evaluate Factual Accuracy in Molecular Comprehension

TL;DR

MoleculeQA is a large, novel dataset designed to evaluate the factual accuracy of molecular understanding in language models through question answering, addressing a critical gap in molecular research evaluation.

Contribution

It introduces the first benchmark and largest QA dataset for assessing factual bias in molecular language models, highlighting their deficiencies.

Findings

Existing models show significant factual inaccuracies in molecular understanding.

MoleculeQA exposes specific weaknesses in current molecular language models.

The dataset provides a new standard for evaluating and improving molecular comprehension in AI.

Abstract

Large language models are playing an increasingly significant role in molecular research, yet existing models often generate erroneous information, posing challenges to accurate molecular comprehension. Traditional evaluation metrics for generated content fail to assess a model's accuracy in molecular understanding. To rectify the absence of factual evaluation, we present MoleculeQA, a novel question answering (QA) dataset which possesses 62K QA pairs over 23K molecules. Each QA pair, composed of a manual question, a positive option and three negative options, has consistent semantics with a molecular description from authoritative molecular corpus. MoleculeQA is not only the first benchmark for molecular factual bias evaluation but also the largest QA dataset for molecular research. A comprehensive evaluation on MoleculeQA for existing molecular LLMs exposes their deficiencies in specific areas and pinpoints several particularly crucial factors for molecular understanding.