Download

The instructions of predicing functions:

  1. Downloading the source code and model weights of ProFun-SOM and the GO Mappings file
  2. Decompressing the source code
  3. Assign the absolute path of the fasta format file of the query protein , the dataset for alignment , the hhblits install directory , model config directory , model weight files (cco, mfo, and bpo), a pickle format file of the go information and a target file for saving the predictions to the parameter fapath , dbpath , hhlib configdir , weightpaths , golabelpath , and saving_path
  4. The MSA_BUFFER_SIZE = 10000, MAXLEN = 2000, TOP_K = 40 in default
  5. In the ubuntu 18.04 LTS, we could set these parameters by environment variables
    • python predict.py ${QUERY} \
    • ${DB} \
    • ${HHLIB} \
    • config/som \
    • ${CONFIG} \
    • ${CCO} \
    • ${MFO} \
    • ${BPO} \
    • ${GO} \
    • ${SAVING}

Project:

ProFun-SOM-source-V1

Trained weights:

CCO model
MFO model
BPO model

Dependency softwares:

Ubuntu 18.04 LTS
Python 3.9
Anaconda3
HMMER3
Pytorch 1.9.1

Benchmark dataset:

CAFA3 data

GO Terms (obo format)

GO Terms

GO Mappings

GO Mappings

Python Requirements

  • biopython==1.79
  • numpy==1.21.5
  • pandas==1.4.2
  • PyYAML==6.0
  • scikit_learn==1.1.2
  • scipy==1.7.3
  • torch==1.9.1