DeepCoil

Fast and accurate prediction of coiled coil domains in protein sequences

New in version 2.0

• Retrained with the updated dataset based on SamCC-Turbo labels.
• Faster inference time by applying SeqVec embeddings instead of psiblast profiles.
• Heptad register prediction (a and d core positions).
• No maximum sequence length limit.
• Convenient interface for using DeepCoil within python scripts.
• Automated peak detection for improved output readability.
• Simplified installation with pip.

Older DeepCoil versions are available here.

Requirements and installation

DeepCoil requires python>=3.7,<3.9 and pip>=19.0.

The most convenient way to install DeepCoil is to use pip:

$ pip3 install deepcoil

Usage

Running DeepCoil standalone version:

deepcoil [-h] -i FILE [-out_path DIR] [-n_cpu NCPU] [--gpu] [--plot]
                [--dpi DPI]

In a rare case of deepcoil being not available in your PATH after installation please look in the $HOME/.local/bin/ or other system specific pip directory.

Description of columns in output file:

• aa - amino acid in the input protein sequence
• cc - sharpened coiled coil propensity
• raw_cc - raw coiled coil propensity
• prob_a - probability of a core position
• prob_d - probability of d core position

Running DeepCoil within script:

from deepcoil import DeepCoil
from deepcoil.utils import plot_preds
from Bio import SeqIO

dc = DeepCoil(use_gpu=True)

inp = {str(entry.id): str(entry.seq) for entry in SeqIO.parse('example/example.fas', 'fasta')}

results = dc.predict(inp)

plot_preds(results['3WPA_1'], out_file='example/example.png')

results[entry] for an entry of sequence length N contains two keys:

• ['cc'] - per residue coiled coil propensity ([N, 1] shape)
• ['hept'] - per residue core positions ([N, 3] shape, order in the second axis is: no/other position, a position, d position)

Peak detection can be performed with the deepcoil.utils.sharpen_preds helper function.

Example graphical output: