Sabari Kumar
Speaker's Institution
Colorado State University
4:00 pm
Chemistry A101
Mixer Time
3:45 pm
Mixer Location
Chemistry B101E
Additional Information

About the Seminar

Despite advances in software algorithms and computer hardware, the simulation of long timescale processes involving biological macromolecules remains prohibitively computationally expensive. Thus, there is a pressing need for novel computational techniques capable of resolving this issue. Harnessing recent advances in neural network-based machine learning algorithms, Noé and coworkers propose an improvement to the previously proposed VAMPNet computational framework (1) for the simulation of long duration events in molecular systems. By learning both a means of dividing up the system studied into independent domains and an approximation of the kinetics of the independent macromolecular domains, the new iVAMPNet architecture extends the utility of VAMPNets, allowing for the study of larger and more complex systems (2). The iVAMPNet architecture is tested on model systems and its accuracy in predicting the timescales of long-duration processes from short molecular dynamics simulations is assessed.


  1. (1) Mardt, A.; Pasquali, L.; Wu, H.; Noé, F. VAMPnets for Deep Learning of Molecular Kinetics. Nat Commun 2018,
    9 (1), 5. https://doi.org/10.1038/s41467-017-02388-1.
  2. Mardt, A.; Hempel, T.; Clementi, C.; Noé, F. Deep Learning to Decompose Macromolecules into Independent
    Markovian Domains. bioRxiv March 31, 2022, p 2022.03.30.486366.
Erica Dawson, Ph.D.