The code in this repository performs 2 tasks from our 3D-AF-Surfer paper (doi: https://doi.org/10.1101/2021.10.21.465371). The protein structure search website for the paper can be found at https://kiharalab.org/3d-surfer/submitalphafold.php

TASK 1: Given two PDB files, predict the structural dis-similarity probability.

This step generates the (dis)-similarity probability for the structure.

Dependencies:

• PyTorch (https://pytorch.org)
• EDTSurf (https://zhanggroup.org/EDTSurf/) [download and place in the ./bin/ directory]
• BioPython (https://biopython.org/)

Usage : run_methods.sh pdb1 pdb2 Example:

./run_methods.sh pdb1 pdb2

The code takes the 2 pdb files as arguments. The generated input feature (3DZD) is stored in the data/ directory The predictions is stored in the output/ directory

TASK 2: Given one or more PDB files, predict the fold classifications based on secondary structure.

This step uses a bagged SVM classifier to predict fold class from secondary structure.

Dependencies:

• DSSP (https://swift.cmbi.umcn.nl/gv/dssp/)
• BioPython (https://biopython.org/)
• scikit-learn (https://scikit-learn.org)

Usage:

python3 predict_fold_class_by_model.py pdb1 pdb2...

Questions/Inquires:

Contact Prof. Daisuke Kihara at [email protected]