Conditionally Generative Methods in Artificial Intelligence Combined to Molecular Simulation for Understanding and Designing Biomolecules

• Grant number: 24/13657-2
• Support Opportunities: Scholarships abroad - Research
• Start date: July 15, 2025
• End date: July 14, 2026
• Field of knowledge: Biological Sciences - Biophysics - Molecular Biophysics
• Principal Investigator: Guilherme Menegon Arantes
• Grantee: Guilherme Menegon Arantes
• Host Investigator: Sergey Ovchinnikov
• Host Institution: Instituto de Química (IQ). Universidade de São Paulo (USP). São Paulo , SP, Brazil
• Institution abroad: Massachusetts Institute of Technology (MIT), United States

Abstract

Artificial intelligence (AI) methods have revolutionized fields such as language processing, image synthesis, and biomolecular structure prediction. Generative AI models have shown great potential in designing biomolecules with specific properties and structures, applicable in drug discovery, enzyme engineering, synthetic biology, and related areas. A major challenge in biomolecular design lies in balancing structural integrity with desired functional properties. Combining AI generative models with physics-based molecular simulation techniques may provide a way towards more efficient and useful design methods, and also create new tools for understanding molecular mechanisms in biophysics.Here, we propose to investigate current approaches to structure-guided multi-objective optimization using generative AI methods such as hallucination and diffusion models. We will explore how information from molecular simulations of chemical reaction paths, obtained with hybrid quantum mechanics/molecular mechanics (QM/MM) potentials, and of conformational sampling, obtained with classical molecular dynamics (MD) trajectories, could be integrated into training and inference workflows to ensure that generated molecules exhibit realistic and stable structures while meeting desired physico-chemical and functional properties.This proposal will be carried out in collaboration with Prof. Sergey Ovchinnikov and his research group in the Department of Biology at the Massachusetts Institute of Technology (MIT), USA, during the proponent's sabbatical visit. Prof. Ovchinnikov is a leader in protein design, with pioneering work in structure prediction using evolutionary information and in hallucination methods. Interaction with his group, the MIT faculty and other research groups, and the research community in the Boston area will be fundamental to the success of this proposal.The proponent, with 20 years of experience in molecular simulation and pioneering work in hybrid QM/MM methods, runs the Computational Biochemistry and Biophysics group at USP. This proposal aims to integrate biomolecular structural prediction and generative AI techniques into the expertise of his group. These methods will be applied in a Thematic Project recently submitted to FAPESP by the proponent. Considering the transformative impact of AI, we anticipate that these proposals will bring significant advancements in scientific research and innovation, with a leading role in AI and overall gains to the Estate of São Paulo.