abs's Introduction

tuned-abs

You can try tuned-abs in Google Colab.

Requirements

The following tools should be installed:

HMMER (jackhmmer and esl-reformat binaries are used)
HH-suite (hhsearch binary is used)

In conda environment, HMMER and HH-suite can be installed with the following commands:

$ conda install bioconda::hmmer
$ conda install -c conda-forge -c bioconda hhsuite

Installation

pip3 install git+https://github.com/bogdanmagometa/abs

Usage

Note: the first use of the pipeline will cause the download of weights unless other pipelines that require the same weights have been already used.

From command line

$ python3 -m tuned_abs --help
usage: python3 -m tuned_abs <input fasta> <output pdb> [--pipeline PIPELINE]

positional arguments:
  input_fasta          Path to .fasta file with two sequences (one for each chain)
  output_pdb           Save the predicted structure to specified .pdb file

optional arguments:
  -h, --help           show this help message and exit
  --pipeline PIPELINE  Which pipeline to use. Can be one of 'Finetuned', 'Finetuned 1x5',
                       'FinetunedValid', 'FinetunedValid 1x5', 'FinetunedValidRefined',
                       'SingleSequence' (default: SingleSequence)
  --quiet              Inference in quiet mode (default: False)
$ python3 -m tuned_abs input.fasta output.pdb --pipeline 'Finetuned 1x5'

From code

from tuned_abs import Pipeline

pipeline = Pipeline.from_name('SingleSequence') # 'Finetuned', 'Finetuned 1x5', 
# 'FinetunedValid', 'FinetunedValid 1x5', 'FinetunedValidRefined', 'SingleSequence'

pdb_string = pipeline.run(
    {
        'H': 'EVQLVESGGGVVQPGRSLRLSCAASGFTFSSYGMHWVRQAPGKGLEWVAVIWYDGSNRYYADSVKGRFTISRDNSKNTLFLQMNSLRAEDTAVYYCHRNYYDSSGPFDYWGQGTLVTVSS',
        'L': 'DIQMTQSPSTLSASVGDRVTITCRASQFISRWLAWYQQKPGKAPKLLIYKASSLESGVPSRFSGSGSETHFTLTISSLQPDDVATYYCQEYTSYGRTFGQGTKVEIKRTV',
    }
)

with open('output.pdb') as f:
    f.write(pdb_string)