DeepUSPS: Deep Learning-Empowered Unconstrained-Structural Protein Sequence Design

This is a kind of model to design unconstrained-structural protein sequences The architecture of the unconstrained-structural protein sequence design model-DeepUSPS

Requirements

# Our version of Cuda is 11.1, version of cuDNN is 8.4.1, and operation system is Ubuntu(20.04)
# Once this is done, you can run the following commands to install the required environment:
conda env create -f environment.yml

Usage

usage: hallucinate.py [-h] [-l LEN] [-s SEQ] [-o FAS] [--ocsv= CSV]
                      [--SPFESN= SPDIR] [--RTIDR= RTDIR] 
                      [--aa_weight= AA_WEIGHT]

optional arguments:
  -h, --help              show this help message and exit
  -l LEN, --len= LEN      sequence length (default: 100)
  -s SEQ, --seq= SEQ      starting sequence (default: )
  -o FAS, --ofas= FAS     save final sequence to a FASTA files (default: )
  --ocsv= CSV             save trajectory to a CSV files (default: )
  --SPFESN= SPDIR         path to SPFESN network weights (default: ../SPSP)
  --RTIDR= RTDIR          path to RTIDR network weights (default: ../RTRT)
  --aa_weight= AA_WEIGHT  weight for the aa composition biasing loss term (default: 0.0)

DeepUSPS a random protein of length 200

python ./DeepUSPS.py -l 200 -o seq.fa

DeepUSPS starting from a given sequence and save trajectory to a CSV files

python ./DeepUSPS.py \
	-s KVFGRCELAAAMKRHGLDNYRGYSLGNWVCAAKFESNFNTQATNRNTDGSTDYGILQINSRWWCNDGRTPGSRNLCNIPCSALLSSDITASVNCAKKIVSDGNGMNAWVAWRNRCKGTDVQAWIRGCRL  \  # PDB ID 6LYZ
	--aa_weight=1.0 \              # turn on amino acid composition biasing term
	--ocsv= seq.csv

Acknowledgement

We refer to the paper of De novo protein design by deep network hallucination. We are grateful for the previous work of I Anishchenko, TM Chidyausiku, S Ovchinnikov, SJ Pellock, D Baker.