Sequential Contrastive and Deep Learning Models to Identify Selective Butyrylcholinesterase Inhibitors

Area of Science:

• Pharmacology
• Computational Chemistry
• Neuroscience

Background:

• Late-stage Alzheimer's Disease (AD) presents challenges for symptomatic treatment.
• Acetylcholinesterase (AChE) inhibitors used in AD can cause harsh side effects.
• Butyrylcholinesterase (BChE) is a potential therapeutic target for AD, with selective inhibitors (BIs) offering a promising alternative.

Purpose of the Study:

• To identify selective BChE inhibitors (BIs) using various machine learning (ML) strategies.
• To optimize ML models for precision in predicting BChE selectivity.
• To compare the efficacy of supervised contrastive learning (CL), deep learning (DL), and Random Forest (RF) for identifying BIs.

Main Methods:

• Employed supervised contrastive learning (CL), deep learning (DL), and Random Forest (RF) ML models.
• Configured models in both single and sequential modeling approaches.
• Virtually screened a library of 5 million compounds using the developed ML models.
• Experimentally validated 20 top predicted compounds in vitro.

Main Results:

• Achieved a 35% hit rate in identifying selective BChE inhibitors from the virtual screen.
• Seven out of 20 tested compounds demonstrated selectivity for BChE over AChE.
• The ML models proved highly efficient in predicting selective BIs.

Conclusions:

• Machine learning, particularly supervised CL, DL, and RF, offers a highly efficient strategy for identifying selective BChE inhibitors.
• This in silico approach can accelerate the discovery of novel therapeutic agents for Alzheimer's Disease.
• The identified selective BIs warrant further investigation for AD symptomatic treatment.