Protein-ligand data at scale to support machine learning

Area of Science:

• Drug Discovery
• Chemical Biology
• Computational Chemistry

Background:

• The Target 2035 initiative seeks to develop pharmacological modulators for every human protein by 2035.
• Small-molecule hit discovery is a critical bottleneck in developing these chemical probes.
• Advancing computational methods is essential to overcome this challenge.

Purpose of the Study:

• To outline the Target 2035 roadmap for enhancing computational methods in small-molecule hit discovery.
• To establish large, open-access datasets of protein-small-molecule binding data.
• To foster machine learning model development for predicting novel binders.

Main Methods:

• Generating high-quality protein-small-molecule binding data using affinity-selection mass spectrometry and DNA-encoded chemical library screening.
• Creating and openly sharing both positive and negative binding datasets.
• Challenging the machine learning community to build predictive models using these datasets.

Main Results:

• Anticipates the identification of experimentally verified hits for thousands of human proteins by 2030.
• Aims to advance open-access algorithms for predicting small-molecule binders.
• Iterative cycles of prediction and testing will refine models and improve prediction success.

Conclusions:

• The Target 2035 roadmap leverages open data and machine learning to accelerate chemical probe discovery.
• This initiative will significantly advance computational approaches for identifying drug-like small molecules.
• The project aims to provide predictive tools for proteins lacking experimental binding data.