Rational Inhibitor Discovery for BRAF^V600E Using PSeMut, a Sequence-Driven Model.

Area of Science:

• Computational chemistry
• Drug discovery
• Genomics

Background:

• Mutations in drug targets reduce small-molecule drug effectiveness and drive disease progression.
• Current methods for discovering mutation-tolerant drugs are not easily scalable for high-throughput screening.
• There is a need for efficient computational tools to design drugs that maintain potency against mutated targets.

Purpose of the Study:

• To introduce PSeMut, a structure-free Siamese model for predicting mutation-induced changes in drug activity.
• To evaluate PSeMut's performance against established methods using a variant-resolved benchmark.
• To demonstrate the utility of PSeMut within a structure-free drug discovery pipeline.

Main Methods:

• Developed PSeMut, a structure-free Siamese model utilizing PSICHIC-derived protein-ligand fingerprints to predict activity changes.
• Benchmarked PSeMut against classical models on variant-resolved data, assessing performance via RMSE.
• Implemented a structure-free prioritization pipeline integrating scaffold novelty, activity scoring, mutation tolerance ranking, and clustering for experimental validation.

Main Results:

• PSeMut achieved a test RMSE of 0.400 ± 0.025, outperforming classical baselines.
• Removing the exchange-consistency constraint in PSeMut led to decreased performance.
• The PSeMut-integrated pipeline successfully prioritized SNS-314, which exhibited mutation-selective cellular activity and suppressed tumor growth in vivo.

Conclusions:

• PSeMut is an effective tool for predicting mutation-induced activity changes in a structure-free manner.
• The developed screening workflow links sequence-based modeling to experimental validation for rational drug design.
• PSeMut enables the prioritization of mutation-resilient drug scaffolds for various diseases.