DynaMine v2, an updated version of the sequence-to-dynamics predictor

Research output: Unpublished contribution to conferencePoster


We present DynaMine v2, the successor of DynaMine[1], a predictive tool for protein backbone dynamics from protein sequence. For DynaMine v2 we processed nuclear magnetic resonance (NMR) chemical shift data into their ShiftCrypt index[2] to define 5 conformational states, a richer representation to better assess the protein conformation and dynam- ics. The molecular dynamics (MD) data was then tagged according to those predefined conformational states, and our definition of general dynamics was calculated from it. This was used as the training data for a neural network (NN) dy- namics regressor. For now, only protein dynamics have been predicted with promising results, and the 5 conformational states defined in this work will be predicted in the next steps for the project. Our network will also internally predict the ShiftCrypt values for each amino acid, although they will not be presented to the user. We expect to improve generalization and therefore improve our predictions when the network is forced to predict conformational states and ShiftCrypt values
Original languageEnglish
Publication statusPublished - 27 Oct 2020
EventBioSB 2020 - Congrescentrum De Werelt, Lunteren, Netherlands
Duration: 27 Oct 202028 Oct 2020


ConferenceBioSB 2020
Internet address


Dive into the research topics of 'DynaMine v2, an updated version of the sequence-to-dynamics predictor'. Together they form a unique fingerprint.

Cite this