F-CPI: A Multimodal Deep Learning Approach for Predicting Compound Bioactivity Changes Induced by Fluorine...

Area of Science:

• Medicinal Chemistry
• Computational Chemistry
• Drug Discovery

Background:

• Fluorine substitution is a key strategy in drug discovery.
• Predicting bioactivity changes from fluorine substitution and its impact on compound-protein interactions (CPI) remains challenging.
• Accurate prediction models for fluorine's effect on CPI are lacking.

Purpose of the Study:

• To develop a predictive model for bioactivity changes induced by fluorine substitution.
• To investigate the impact of fluorine substitution on compound-protein interactions (CPI).
• To enhance the efficiency of drug discovery and design through accurate prediction.

Main Methods:

• Construction of a large dataset comprising 111,168 pairs of fluorine-substituted and non-fluorine-substituted compounds.
• Development of a multimodal deep learning model named F-CPI.
• Comparative analysis of F-CPI against traditional machine learning and existing CPI models.

Main Results:

• The F-CPI model demonstrated superior performance with accuracy (∼90%), precision (∼79%), and recall (∼45%) compared to GraphDTA.
• F-CPI application led to over a 100-fold increase in bioactivity for SARS-CoV-2 3CLpro inhibitors.
• Optimized compounds showed significantly improved inhibitory concentration 50 (IC50) values (0.23 μM vs 28.19 μM).

Conclusions:

• The multimodal deep learning model F-CPI is an effective tool for predicting fluorine substitution effects on bioactivity.
• F-CPI facilitates structural optimization in drug discovery, leading to significant bioactivity enhancements.
• This model offers a valuable approach for advancing drug design and development processes.