ligand-discovery.ai: Advancing Chemoproteomics Research with AI-Powered Web Tools

ligand-discovery.ai is a cutting-edge research initiative from Georg Winter's group at CeMM, offering a suite of powerful web applications designed to explore and analyze fragment-based chemoproteomics screening data. Tailored for researchers in drug discovery and molecular biology, the platform enables deep insights into ligand-protein interactions through advanced machine learning models and interactive data visualization. With five core tools—Interactions, Explore protein sets, Protein set enrichment analysis, Fragment predictor, and On-the-fly modeling—it supports both hypothesis generation and data-driven validation. The Fragment predictor leverages 12 high-performing models (3 Global, 9 Specific, and 10 Ontology-based) with AUROC > 0.75, ensuring reliable predictions of promiscuous or specific interactome signatures. Data thresholds (10% for proteins, 5% for fragments) help classify interaction patterns accurately. All tools are accessible via cloud-hosted web apps or can be run locally by forking open-source repositories, with screening and competition data available for download to fuel independent research. Built on Python and Jupyter Notebook with transparent licensing under GPL-3.0 and AGPL-3.0, the platform combines scientific rigor with user-friendly design.

Key Features:

• Interactions App: Explore ligand-protein interactions across 407 fragments with competition assay visualization.
• Explore Protein Sets: Input protein sets to identify fragment interactions and classify them by promiscuity or specificity.
• Protein Set Enrichment Analysis: Analyze fragment profiles using detailed protein annotations (domains, families, functions, localization).
• Fragment Predictor: Predict promiscuity vs. specificity of fully-functionalized fragments using 12 validated models (AUROC > 0.75 considered strong).
• On-the-Fly Modeling: Build custom machine learning models to predict fragment-protein interactions using organized protein subsets.
• Threshold-Based Classification: Uses 10% hit/regulated ratio for proteins and 5% for fragments to define promiscuity.
• Open-Source & Local Deployment: All tools available via GitHub repositories, runnable locally after forking.
• Data Accessibility: Screening and competition data available for download to support independent research.
• Scientific Validation: Backed by peer-reviewed research (Offensperger et al., Science 2024) and developed by leading AI and biomedical scientists.

Pricing: ligand-discovery.ai is offered as a free, open-access research resource with no cost to users.

Conclusion: ligand-discovery.ai is an exceptional, freely available AI-driven platform that empowers researchers to unlock new insights in chemoproteomics, combining robust predictive modeling with intuitive tools for data exploration and analysis.