A general prediction model for compound-protein interactions based on deep learning.

Area of Science:

• Computational biology
• Pharmacology
• Bioinformatics

Background:

• Compound-protein interactions (CPIs) are vital for drug discovery and understanding drug mechanisms.
• Existing computational methods struggle with accuracy and generalization due to data limitations and diverse compounds/targets.
• Traditional Chinese Medicine (TCM) offers a rich source of compounds, but identifying their targets and interactions remains challenging.

Purpose of the Study:

• To develop a robust computational model for accurate CPI prediction using deep learning.
• To apply the model to identify targets of TCM compounds and predict synergistic drug combinations.
• To experimentally validate the predicted synergistic drug combinations for potential therapeutic applications.

Main Methods:

• Developed a deep learning (DL) model integrating a large-scale bioactivity dataset for CPI prediction.
• Applied the DL model to predict targets of active compounds from the TCM herb pair *Astragalus membranaceus* and *Hedyotis diffusa*.
• Used predicted targets to identify synergistic multi-compound combinations via the DeepMDS model and validated through *in vitro* assays.

Main Results:

• The DL model achieved high performance with an AUROC of 0.98, AUPR of 0.98, and 93.31% accuracy.
• Successfully predicted an expanded dataset of compound targets from the studied TCM herb pair.
• *In vitro* experiments confirmed the efficacy of predicted synergistic combinations against breast cancer cells (MDA-MB-231).

Conclusions:

• The developed CPI prediction model demonstrates superior accuracy and generalization capabilities.
• The model facilitates the discovery of novel CPIs and synergistic drug combinations, particularly from TCM.
• This approach shows significant potential for accelerating drug discovery and elucidating TCM's bioactive compounds.