TransMut: Automatically optimize mutated peptide program from the Transformer-derived self-attention model to predict peptide and HLA binding

webserver: https://issubmission.sjtu.edu.cn/TransPHLA-AOMP/index.html

1. All software dependencies and oprating system:

The dependencies is the pytorch environment on linux system, the operating system is CentOS Linux release 7.7.1908.

2. Any required non-standard hardware: The GPU we used GeForce RTX 3080 GPU and CUDA 11.1.

3. Software version: This is the 1th version, this verision has been tested on indenpendent set, external set, neoantigen data, and HPV vaccine data.

4. Installation guide:

(1) Instructions:

Install pytorch environment on your linux system: pip install -r pHLAIformer_simple_requirements.txt pHLAIformer_simple_requirements.txt is a simplified version of the dependency package selected by the author. If the dependency package is not installed enough, you can check pHLAIformer_requirements.txt to check the version of the corresponding dependency package. The installation of pytorch needs to correspond to your system, GPU and other versions, please check https://pytorch.org/get-started/locally/, we are using GeForce RTX 3080 GPU and CUDA 11.1.
Enter the current directory: cd ./TransPHLA-AOMP/

(2) Typical install time on a "normal" destop computer: The install time for this code cannot be determinated, it depends on the network situation and computer configuration. The user-friendly webserver is no need to be installed, so there is no install time.

5. Demo:

(1) Instructions to run on data:

Code: python pHLAIformer.py --peptide_file "peptides.fasta" --HLA_file "hlas.fasta" --threshold 0.5 --cut_length 10 --cut_peptide True --output_dir './results/' --output_attention True --output_heatmap True --output_mutation True
Webserver: click the Example button on the https://issubmission.sjtu.edu.cn/TransPHLA-AOMP/index.html and click the submit button.

(2) Excepted output: The output can be seen in https://issubmission.sjtu.edu.cn/TransPHLA-AOMP/tutorials.html

(3) Expected run time for demo on a "normal" desktop computer:

Code: The prediction of 170000 peptide-HLA binders only needs 2 minutes on a CPU.
Webserver: For one sample, if we choose to generate all optional output, we need approximately 30

6. Instructions for use:

(1) How to run the software on your data: https://issubmission.sjtu.edu.cn/TransPHLA-AOMP/tutorials.html

(2) Reproduction instructions: You can input the same data many times on the webserver to obtain the reproduction results.

peptides.fasta and hlas.fasta are sample files provided by us, which contain some error debug cases.

Parameter:

peptide_file: type = str, help = the filename of the .fasta file contains peptides
HLA_file: type = str, help = the filename of the .fasta file contains sequence
threshold: type = float, default = 0.5, help = the threshold to define predicted binder, float from 0 - 1, the recommended value is 0.5
cut_peptide: type = bool, default = True, help = Whether to split peptides larger than cut_length?
cut_length: type = int, default = 9, help = if there is a peptide sequence length > 15, we will segment the peptide according the length you choose, from 8 - 15
output_dir: type = str, help = The directory where the output results are stored.
output_attention, type = bool, default = True, help = Output the mutual influence of peptide and HLA on the binding?
output_heatmap: type = bool, default = True, help = Visualize the mutual influence of peptide and HLA on the binding?
output_mutation: type = bool, default = True, help = Whether to perform mutations with better affinity for each sample?

cloudaner115/TransPHLA-AOMP