Evaluating molecular fingerprint-based models of drug side effects against a statistical control. Issue 11 (November 2022)

• Record Type: Journal Article
• Authors: Alpay, Berk A.; Gosink, Mark; Aguiar, Derek
• Is Part Of: Drug discovery today. Volume 27:Issue 11(2022)
• Publication Date: 2022-11
• Subjects: Side effects -- Machine learning -- Benchmarking -- Chemical structure; Drugs -- Design -- Periodicals; Drugs -- Research -- Periodicals; 615.1
• DOI: 10.1016/j.drudis.2022.103364
• Languages: English
• ISSNs: 1359-6446

Abstract

Highlights: Side effects appear well-predicted using only their frequencies. Baselines for common metrics depend on the distribution of side effect frequencies. Molecular fingerprints help rank drugs by likelihood to cause a given side effect. Molecular fingerprints help marginally when ranking the side effect likelihoods of one or more drugs. Fingerprint-based models are at most modestly skilled overall at identifying side effects associated with drugs. Abstract : There are many machine learning models that use molecular fingerprints of drugs to predict side effects. Characterizing their skill is necessary for understanding their usefulness in pharmaceutical development. Here, we analyze a statistical control of side effect prediction skill, develop a pipeline for benchmarking models, and evaluate how well existing models predict side effects identified in pharmaceutical documentation. We demonstrate that molecular fingerprints are useful for ranking drugs by their likelihood to cause a given side effect. However, the predictions for one or more drugs overall benefit only marginally from molecular fingerprints when ranking the likelihoods of many possible side effects, and display at most modest overall skill at identifying the side effects that do and do not occur.