Relating Essential Proteins to Drug Side-Effects Using Canonical Component Analysis: A Structure-Based Approach.

Area of Science:

• Pharmacology
• Computational Biology
• Drug Discovery

Background:

• The molecular basis of many drug side-effects remains unknown, hindering prediction and patient safety.
• Existing methods primarily focus on established drug targets, potentially overlooking off-target effects.
• Low-affinity binding to non-traditional targets may significantly contribute to adverse drug reactions.

Purpose of the Study:

• To develop a computational method for predicting drug side-effects by considering interactions with essential human proteins.
• To identify novel drug-target associations that may explain previously unknown side-effect mechanisms.
• To assess the role of non-canonical protein targets in driving drug-induced adverse events.

Main Methods:

• Systematic prediction of drug binding affinities to 563 essential human proteins using their 3D structures.
• Benchmarking affinity predictions against available experimental data.
• Application of singular value decomposition and canonical component analysis (SVD-CCA) to correlate binding profiles with drug side-effects.

Main Results:

• Achieved good accuracy in predicting drug side-effects (average AUC: 0.82 for rare side-effects).
• Identified 2768 novel associations between 50 essential proteins, 99 drugs, and 77 side-effects.
• Demonstrated that side-effect frequency can confound mechanistic studies, and the method isolates novel biological signals.

Conclusions:

• Focusing on essential proteins uncovers potential drug side-effect associations missed by traditional target-based approaches.
• The findings provide new insights into the molecular mechanisms underlying drug side-effects.
• Highlights the necessity for broader experimental investigation of drug binding across the human proteome.