Ultrafast Protein Dynamics by Machine Learning

To date, significant efforts have been made to understand the structural dynamics of biological molecules, nevertheless, a precise description of them remains an unsolved problem in biophysics. The answer to this question helps explain many biological phenomena and the creation and work cycle of living organisms. It will also provide revolutionary and new applications and products in other fields such as pharmacology, medicine, biotechnology, material science, and so on. The research in this area most commonly relies on a few experimental techniques, such as X-ray crystallography, cryo-electron microscopy, and NMR spectroscopy. However, the level of information extracted by conventional data analysis algorithms is severely restricted by data artifacts such as noise, timing inaccuracy, and data incompleteness. Therefore, access to many intrinsic molecular dynamics that occur in sub-picosecond timescales remains difficult. To tackle this problem, we have developed and validated a data analytic machine learning method capable of extracting detailed information about the dynamics of biomolecules from experimental data. In this presentation, I will review the method and show some key results from the ultrafast structural dynamics of a photosensitive protein.