Machine learning models for pharmacogenomic variant effect predictions - recent developments and future frontiers.

Area of Science:

• Genomics
• Pharmacogenomics
• Computational Biology

Background:

• Genetic variations significantly influence individual drug responses and toxicity.
• Millions of rare genetic variants remain functionally uncharacterized, hindering precision medicine.
• Machine learning (ML) offers advanced methods for predicting variant effects.

Purpose of the Study:

• To review current ML-based variant effect predictors for pharmacogenomics.
• To discuss methodological differences, strengths, and limitations of these tools.
• To explore emerging methods for predicting substrate-specificity and epistasis.

Main Methods:

• Utilizing DNA and protein sequences, evolutionary conservation, and haplotype structures.
• Employing deep learning models for evolutionary conservation and biophysical properties.
• Integrating multiple predictive models through ensemble approaches for enhanced accuracy.

Main Results:

• ML models significantly improve the prediction of variant effects on drug response.
• Deep learning and ensemble methods show increased accuracy, robustness, and interpretability.
• Emerging methodologies address substrate-specificity and variant epistasis.

Conclusions:

• ML-based tools are crucial for functionally characterizing drug-related genetic variants.
• These predictors offer a viable strategy to translate genomic data into pharmacogenetic recommendations.
• Advancements in ML are key to realizing the full potential of precision medicine.