aemcmc

MCMC simulations of proteins using autoencoder-generated structures

In this repository, you can find jupyter notebooks with the code used to perform MCMC simulations of (mini)proteins.

• Notebook trj2npy contains code used to prepare training dataset - extract Cartesian coordinates of protein atoms from supplied xtc file and save it as a numpy array.
• Notebook ae contains the code for building the autoencoder ML model, trains it on the (mini)protein structures and saves it.
• Notebook analyse_ls contains code for analysing the AE latent space, and allown for precise selection of latent space coordinates, decoding a protein structure from them and calculating its potential energy.
• Notebook MC provides the code for loading the trained model, conducting MCMC simulation using the decoder, and analysing the sampled population of structures.

Notebooks analyse_ls and MC can be ran straight after the ML model is prepared in ae in any order.