ChemSpace Copilot: Agentic AI for Interactive Visualization and Exploration of Chemical Space
Year: 2026
Type: preprint
Abstract: Chemography provides a foundational approach to exploring chemical space by compressing high-dimensional representations of chemical structures into 2D maps, enabling clearer interpretation of similarity, diversity, and activity. Such maps power core chemoinformatics tasksstructure-activity relationship analysis, library design, virtual screening, and de novo design. In practice, effective use of chemography relies on manual application of data analysis tools and expert-driven interpretation of the resulting landscapes for hypothesis generation. To automate chemography-guided discovery, we introduce ChemSpace Copilotan agentic AI application that encapsulates standard chemographic workflows based on the non-linear dimensionality reduction technique Generative Topographic Mapping (GTM). ChemSpace Copilot enables researchers to perform chemographic tasks end-to-end through an intuitive, chatbot-style interface. Its modular toolkit integrates chemical data retrieval from ChEMBL, GTM parameter optimization, chemotype analysis and report generation. We further demonstrate the extensibility of the framework by integrating generative deep learning models with the synthesis planning, enabling effective and illustrative end-to-end molecular discovery from design to synthetic feasibility. We demonstrate its utility across three tasks: exploration of ChEMBL's chemical space, chemography-guided compound generation, and synthesis prediction.
Source: ChemRxiv
Institution: Université de Strasbourg
Language: English
Subfield: Drug Discovery
Domain: Life Sciences
Open Access status: green
OpenAlex ID: W7133232155