Feed-Forward Deep Neural Networks Predict Substrate-Specific Effects of Transporter Variants to Explain Drug Response Variability | JoVE Visualize

Area of Science:

Pharmacogenomics
Computational Biology
Biochemistry

Background:

Genetic variations in drug transporter genes contribute to individual differences in drug efficacy and toxicity.
Current predictive models for transporter variants are substrate-agnostic, limiting their ability to capture drug-specific effects.
Understanding variant function is crucial for personalized medicine and optimizing drug therapy.

Purpose of the Study:

To develop and validate a novel computational model, the substrate-specific effect predictor (SSEP), for predicting the functional impact of drug transporter variants.
To enable substrate-dependent prediction of variant effects, addressing the limitations of existing substrate-agnostic tools.
To apply SSEP to large-scale population data for identifying clinically relevant transporter variants.

Main Methods:

Integration of in vitro uptake assays and deep mutational scanning data.
Development of a feed-forward deep neural network architecture leveraging multiscale features from modeled variant-substrate complexes.
Benchmarking SSEP against substrate-agnostic predictors using experimental uptake data across multiple transporter families and substrate classes.

Main Results:

SSEP achieved significant correlation (Spearman's ρ = 0.64) with experimental uptake data across diverse transporter families and substrates.
SSEP demonstrated superior concordance with experimental data compared to substrate-agnostic predictors (Kruskal-Wallis p < 10⁻⁴¹).
In UK Biobank data, OCT1 variant burden, weighted by metformin-specific SSEP scores, was significantly associated with metformin maintenance dose, unlike substrate-agnostic weights.

Conclusions:

SSEP provides quantitative, drug-specific predictions of transporter variant effects, outperforming existing methods.
This substrate-specific approach enhances the identification of clinically significant transporter variants in large datasets.
SSEP has the potential to improve drug response prediction and guide personalized pharmacotherapy.