Developing lipid nanoparticles for genetic medicines is often limited by the lack of historical data, reducing the effectiveness of traditional machine learning. LUMI-lab, an autonomous, self-driving laboratory powered by a 3D transformer-based molecular model, helps overcome this challenge by learning efficiently from small datasets.

Pretrained on more than 28 million molecules, the model works with automated robotics to run continuous active learning cycles that balance high-value predictions with uncertainty-driven exploration. Through ten iterations and more than 1,700 robotically synthetized lipids, the system identified a highly potent, unexpected structural motif — brominated lipid tails — without human bias.

This presentation will cover the system’s computational architecture, uncertainty-aware sampling strategies and how combining foundation models with closed-loop robotics accelerates molecular discovery.

The speaker for this seminar is Yue Xu, Ph.D., an instructor in the Yong Li Lab in the Section of Epidemiology and Population Sciences in the Department of Medicine at Baylor College of Medicine.