CAREER: Multimodal Generative AI for Autonomous Biomolecular Design and Accelerated Discovery

The rational design of new biomolecules, such as enzymes and therapeutic proteins, is essential for advancing medicine and creating sustainable chemical processes. However, conventional frameworks for discovering these molecules remain slow, fragmented, and largely rely on trial-and-error. This project promotes the progress of science by creating a new class of artificial intelligence (AI) models that can design complex biological molecules with unprecedented precision and speed. Specifically, this research will build an autonomous discovery engine that unifies the design of molecular sequences and their three-dimensional shapes into a single, self-improving system that can efficiently design molecules, learn from real-world experimental feedback, and continuously improve its own performance. By shifting the scientific process from manual orchestration to automated AI-driven discovery, this research will significantly reduce the time and cost required to develop life-saving therapeutics and sustainable biochemicals, thereby advancing the national health and prosperity. The project also contributes to the national interest by strengthening the biotechnology workforce through interdisciplinary training for students and public outreach programs that introduce young learners to the intersection of computer science and biology. Finally, all AI models, datasets, and software developed during this project will be shared publicly to broadly enhance the national research infrastructure. The technical goal of this project is to develop a unified, multimodal, and self-improving artificial intelligence framework for high-fidelity biomolecule sequence and structure co-design, alongside iterative optimization. The research activities are organized into three synergistic aims. First, the investigator will develop a novel generative paradigm that natively supports multimodal variable-length co-design through evolution-inspired edit operations, coupled with multi-scale geometric representations and fast-sampling consistency models. Second, the project will create a suite of data-efficient optimization algorithms to intelligently steer and improve these generative models. This includes a unified guided generation framework based on stochastic optimal control, uncertainty-aware sequential Monte-Carlo sampling, and a co-evolutionary reinforcement learning framework that allows the system to continuously improve from experimental feedback. Finally, these computational methods will be integratedinto a fully autonomous design, build, test, and learn cycle. This closed-loop platform will be deployed to solve high-impact biochemical engineering challenges, specifically targeting the design of noncanonical cyclic peptides, new-to-nature enzymes, and high-affinity antibodies. The potential contribution of this work is a fundamental shift from static, fragmented pipelines to an autonomous discovery engine capable of rapidly creating highly functional biomolecules. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria. NSF Award ID: 2543543 | Program: 01002627DB NSF RESEARCH & RELATED ACTIVIT,01002930DB NSF RESEARCH & RELATED ACTIVIT,01003031DB NSF RESEARCH & RELATED ACTIVIT | Principal Investigator: Ge Liu | Institution: University of Illinois at Urbana-Champaign, URBANA, IL | Award Amount: $530,625 View on NSF Award Search: https://www.nsf.gov/awardsearch/show-award/?AWD_ID=2543543 View on Research.gov: https://www.research.gov/awardapi-service/v1/awards/2543543.html