Hierarchical multi-omics data integration and modeling predict cell-specific chemical proteomics and drug responses

Area of Science:

• Computational biology
• Pharmacology
• Systems biology

Background:

• Drug-induced phenotypes arise from complex biomolecular interactions.
• Characterizing these requires integrating multi-omics data, but proteomics data is often scarce and has missing values.
• Existing methods have not fully exploited proteomics for understanding drug actions.

Purpose of the Study:

• To develop a computational method for inferring drug-induced proteome patterns.
• To predict proteome profiles and phenotypes for uncharacterized cells/tissues exposed to chemicals.
• To enable progress in systems pharmacology by addressing proteomics data challenges.

Main Methods:

• Developed TransPro, an end-to-end deep learning framework.
• Hierarchically integrated multi-omics data following the central dogma of molecular biology.
• Applied the framework to predict anti-cancer drug sensitivity and adverse drug reactions.

Main Results:

• TransPro accurately predicts proteome profiles and drug-induced phenotypes.
• The framework's accuracy in predicting drug sensitivity and adverse reactions is comparable to experimental data.
• Demonstrated the utility of TransPro in systems pharmacology applications.

Conclusions:

• TransPro facilitates the imputation of missing proteomics data.
• The framework can aid in compound screening and understanding drug mechanisms.
• TransPro advances systems pharmacology by integrating multi-omics data for drug effect prediction.