This is an individual project which aims to predict polymer structures given the trajectory data of the polymer for a given time. Two research papers were implemented to achieve the required result.
The first research paper used a Variational Autoencoders Network to find an initial Reaction Coordinate and iteratively run biased MD simulations to find the optimal Reaction Coordinate.
The second research paper used an Autoencoder model with a Mixture Density Network to provide a reduced-order representation and then an LSTM-MDN model is used to forecast the latent space dynamics to maximise the latent data likelihood. The high dimensional state configurations are then obtained from the decoder.
Points to be noted: The .db files and the csv files used in the codes have not been uploaded due to size limits. This is a highly unexplored field, therefore results are not perfect, still working on it :).