Accurate Prediction of Protein Structural Flexibility by Deep Learning Integrating Intricate Atomic Structures and Cryo-EM Density Information

Here is an official implementation for RMSF-net. (The current code is a temporary version which will be added more complete features in the future.)

This script is mainly for CPU version, which can be adapted to general computing devices while maintaining a relatively fast processing speed.

System requirements

• Ubuntu 18.04 (recommend, any compatible system is OK.)

Install

1. Downlaod and install UCSF Chimera from https://www.cgl.ucsf.edu/chimera/download.html, and set environment variable "CHIMERA_PATH" as the path to chimera exeutable file, e.g., "path_to_chimera/bin/chimera".

2. Install Anaconda or Miniconda.

3. Run ./install.sh, to create a python environment called rmsf-net.

Typical installation time is expected to be less than 30 minutes.

Usage

For flexibility prediction, execute command:

conda activate rmsf-net
python predict.py -p pdb_file -e emd_file -o output_dir  -m mode 

• pdb_file: The path to PDB file in .pdb format.
• emd_file: The path to cryo-EM map in .map or .mrc format.
• output_dir: The path to save the middle data file generated by the program and the final prediction output.
• mode: The mode selected by the user, which can be 1, 2, or 3, corresponding to the three options "Only PDB model","Only Cryo-EM","Cryo-EM plus PDB model"

We recommened to run this script for your protein of interest with its corresponding cryo-EM map (2-4 Å) and PDB file on mode 3, which is the main method in our work.

Example:

python predict.py -p "data/6FBV.pdb" -e "data/emd_4230.map" -o "results" -m 3 

If you can not set chimera environment variable , you can also specify the chimera exeutable path at the command line, as following:

python predict.py -p "data/6FBV.pdb" -e "data/emd_4230.map" -o "results" -m 3 -ch "path_to_chimera/bin/chimera"

Typical prediction on CPUs will cost less than 3 minutes.

Output

Run the command line above, the program will generate PDB simulated map (sim_map.mrc) and data file (data.pth) and save to "./results" directory.

The predicted RMSF will be normalized and mapped onto the bfactor column of the original PDB file, saved to "./results" directory according to the mode, e.g. "/results/pdb_plus_cryoem/rmsf_nor.pdb" for mode 3.

You can visualize the RMSF output using the output pdb file at UCSF Chimera or PyMOL.

Web-Fronted RMSF-net software

The web-fronted RMSF-net with user-friendly interface can be downloaded from Figshare (https://doi.org/10.6084/m9.figshare.25893670).

The detailed description of this software and its interface are provided in the section "Accessible and user-friendly RMSF-net" of the paper's Supplementary Material.

RMSF annotated pdb files

The datasets containing MD simulation annotations for RMSF-net training and evaluation are accessible for download from Figshare (https://doi.org/10.6084/m9.figshare.25893670).