Speaker: Zhaocheng Liu
Title: Learning protein transition states from simulation data via Flow Matching
Abstract: Proteins are biomolecules responsible for most functions in living organisms. For example, enzymes aid in digestion, and haemoglobin transports oxygen in our blood cells. Most proteins perform their function by switching between different 'states' or configurations. Thanks to decades of experiments and measurements, we now have access to large databases containing snapshots of these states. However, understanding how proteins switch between states remains a challenging problem. Many transition pathways are transient, rare, and difficult to observe or simulate directly. Knowledge of these pathways can be extremely valuable, for instance, in designing targeted drugs against antibiotic-resistant bacteria.
In my work, I study a Periplasmic Binding Protein (PBP) involved in nutrient delivery within Gram-negative bacteria. This protein is known to switch between two major states. Here, we compare several popular computational methods for capturing such transitions, highlighting their strengths and limitations. We then propose a new pipeline that combines machine learning techniques with Molecular Dynamics simulations to identify transition states more effectively, bypassing some of the key limitations in existing approaches.