PconsC4:

Fast, accurate, and hassle-free contact prediction.

Installation instructions:

pip3 install numpy Cython pythran &&
pip3 install pconsc4

NB: the trained model is a bit over Github's limit, so they cannot be checked in the repo. If you need them, you can grab the trained models from the releases tab.

You will also need a deep learning backend compatible with Keras. We recommend Tensorflow:

pip3 install -U tensorflow

Versions

These versions are known to work

keras==2.2.4 2.0< tensorflow >=1.12. pythran 0.9.5

Later versions (such as tensorflow 2) are known to not work.

Usage instructions

Inside Python:

import pconsc4

model = pconsc4.get_pconsc4()

pred_1 = pconsc4.predict(model, 'path/to/alignment1')
pred_2 = pconsc4.predict(model, 'path/to/alignment2')

# Show pred_1 on the screen:

import matplotlib.pyplot as plt 
plt.imshow(pred_1['cmap'])
plt.show()

The program accepts alignments in .fasta, .a3m, or .aln, without line wrapping. The query sequence must be the first line, and it cannot contain gaps.

We also provide a function to format the output in CASP format:

# Save in CASP format:
from pconsc4.utils import format_contacts_casp
print(format_contacts_casp(pred_2['cmap'], seq_2, min_sep=5))

and Cameo:

# Save in Cameo format:
from pconsc4.utils import format_contacts_cameo
print(format_contacts_cameo(pred_2['cmap'], seq_2, min_sep=5))

Troubleshooting:

• pyGaussDCA fails with "template errors":

  Update your compiler. GCC 5 or higher is known to work.

• I get a Segmentation Fault when making predictions

  Ensure the first sequence in your MSA does not contain gaps. If it does, remove the corresponding columns.