Combining chemoproteomics with machine learning identifies functionally active covalent fragments for hard-to-drug cancer drivers

Abstract a) Many cancer drivers are considered “undruggable” and without targeted treatments because they lack binding sites for conventional small molecules. Here, we introduce The FRONTIER™ Platform applying machine learning (ML), chemoproteomics and covalent chemistry to identify binding sites and cell-active covalent fragments across the human proteome, including against most cancer drivers and previously “undruggable” targets. Molecules tested in functional assays are active and can serve as starting points for new drug discovery initiatives. b) We are using mass spectrometry and data analysis workflows to perform high-throughput chemoproteomic profiling experiments. These experiments identify hits across the proteome using different cancer-relevant cell backgrounds and characterize binding sites for drug discovery. Customized ML algorithms using chemoproteomic, genomic and structural data to characterize and prioritize identified binding sites for covalent drug discovery. The performance of the platform allows the profiling of thousands of compounds from a custom-built library that has been optimized by ML for covalent fragment-based drug discovery. The nature of the fragment hits and the ability to map them to and focus on preferred binding sites for covalent drugs enables accelerated lead generation. c) We show details of the platform highlighting library design concepts and the hotspot map with binding site prioritization algorithms. We show coverage and applicability across important cancer target classes and signaling pathways. Covalent fragment hits have been identified for multiple difficult cancer targets, including KRAS, p53, STAT3, KEAP1, PTPN11 and others. The platform also identifies ligands for novel allosteric binding sites in established oncology targets such as CDK4, PI3KCA and BTK. We will highlight how discovered ligands show functional activity in orthogonal assays, demonstrating their fitness for drug discovery campaigns and target validation experiments. d) Undruggable targets across a variety of cancer target classes have become druggable. Citation Format: Johannes C. Hermann, Robert Everley, Laura Marholz, Matthew Berberich, Tzu-Yi Yang, Yu-Hsin Chao, Abduselam K. Awol, Michael Shaghafi, Han Yoon, Rohan Varma, Reed Stein, Karsten Krug, Emily Lachtara, Daniel Erlanson, Chris Varma, Kevin R. Webster. Combining chemoproteomics with machine learning identifies functionally active covalent fragments for hard-to-drug cancer drivers. [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2023; Part 1 (Regular and Invited Abstracts); 2023 Apr 14-19; Orlando, FL. Philadelphia (PA): AACR; Cancer Res 2023;83(7_Suppl):Abstract nr 5333.