Illuminating the Druggable Human Proteome with an AI Protein Profiling Platform

Area of Science:

• Computational biology
• Proteomics
• Drug discovery

Background:

• Proteomic approaches face limitations in coverage and data consistency.
• Existing machine learning models struggle with structural dependencies and heterogeneous data.
• A comprehensive ligandable atlas is crucial for understanding protein function and accelerating therapeutic development.

Purpose of the Study:

• To develop a multimodal AI platform (AiPP) for predicting and characterizing protein ligand interaction sites directly from amino acid sequences.
• To leverage large language models (LLMs) and harmonized databases for improved accuracy and coverage.
• To create a proteome-wide covalent ligandability atlas.

Main Methods:

• Developed AiPP, an AI platform powered by evolutionary-scale protein LLMs.
• Utilized two harmonized machine learning training sets from activity-based protein profiling (ABPP) and co-crystal structures.
• Implemented a LLM representation-based clustering framework to reconcile and augment experimental data.
• Employed complementary protocols for iterative data expansion and model performance improvement.

Main Results:

• AiPP accurately predicts cysteine liganding events, achieving 84% AUPRC and 89% AUROC, even when trained solely on ABPP data.
• Identified consistently and heterogeneously liganded cysteines across cancer cell lines and dynamic, ligandable pockets in "undruggable" transcription factors.
• Predicted previously undetected active-site and allosteric cysteines in protein tyrosine phosphatases.
• Generated a proteome-wide atlas, identifying novel ligandable sites in proteins like MC3R, a target for eating disorders and obesity.

Conclusions:

• AiPP offers a powerful approach to creating a proteome-wide covalent ligandability atlas.
• The platform significantly enhances the identification of ligandable sites, including for previously intractable targets.
• The LLM-based methodology is broadly applicable to large-scale, heterogeneous biological data, advancing proteomics and ML model development.