pyconsFold:

A python modelling framework built on top of CNS. Support for both trRosetta distance predictions and CASP format contact predictions, both binary and distance based.

Install

pyconsFold require a working installation of CNS. This needs to be done manually due to license.

Install CNS
1. Request a download link from CNS.
2. Follow the emailed instructions to download "cns_solve_1.3_all_intel-mac_linux.tar.gz
3. Extract the files tar xzvf cns_solve_1.3_all_intel-mac_linux.tar.gz
4. Change into the resulting directory cd cns_solve_1.3
5. Unhide the bash-specific file mv .cns_solve_env_sh cns_solve_env.sh
6. In this resulting file, replace _CNSsolve_location_ with the CNS installation folder. If you extracted the file in your homefolder then the CNS installation would be: /home/<your username>/cns_solve_1.3
7. Source CNS, source cns_solve_env.sh*, to make this permanent and to prevent you having to do this every time, add it to your .bashrc file.
8. Test CNS by going into the test folder cd test and run the tests ../bin/run_tests -tidy *.inp * If you get an error about csh interpreter, you need to install csh
Install pyconsFold
1. Run pip install pyconsFold
Optional: If you clone this github repo, you can run a suite of tests using python3 run_test.py

Usage

import pyconsFold

pyconsFold.model_dist(fasta, contacts, out_dir)

Functions

model      -- Classic modelling using binary contact predictions (although the contact file can contain distance and errors they wont be used)
model_dist -- Model using distance and errors, requires either a CASP-formated rr file with additional column with standard error in Ångströms or a trRosetta-contacts file in npz-format.
model_dock -- Perform modelling and docking of two protein chains. Requires _one_ contacts file with both inter- and intra-contacts.

Top arguments

rr_pthres	--	Threshold for the confidence we want in a prediction (default model(0.80), model_dist(0.45), model_dock(0.50))
rr_sep		--	Separation between contacts (default 0)
save_step	--	Save working steps (default False)
stage2		--	Run stage2, filter contacts vs generated structure and generate new structures with filtered contacts (default False)
debug		--	Write out debug information (default False)
selectrr	--	How many contacts to use? Can be "all", "#L", or #. (default "all")
mcount		--	How many models to generate? (default 20)
top_models	--	How many of the generated models should be ranked and saved? (default 20)
use_angles	--	If predicted angels should be used, only works with npz (default False)
omega		--	RR-formated file with omega angles (if npz are not used) (default '')
theta		--	RR-formated file with theta angles (if npz are not used) (default '')

Utilities

QA-function arguments to all above functions:

pcons (default False) -- If set to true, gives pcons score for all models (using either pcons installed in the PATH or the builtin binary)
tmscore_pdb_file -- If a structure file is supplied, runs all models against this (presumed) native structure and reports the TMscore (using either TMscore in the PATH or builtin binary)

Extras

from pyconsFold.utils import npz_to_casp, pdb_to_npz

npz_to_casp("trRosetta.npz")  ##  Converts trRosetta distance and angle predictions to CASP format in separate files

pdb_to_npz("structure.pdb")   ##  Converts a structure (pdb/mmCif) to trRosetta distances and angles, useful when investigating how well a model conforms to restraints

Adjustable parameters for CNS, advanced

rrtype		--	Between which atoms in a residue are the contacts? (default 'cb')
lbd			--	Lambda, 0.1-10 (default 0.4)
contwt		--	Contact restraint weights, 0.1-10000 (default 10)
sswt		--	Secondary structure weights, 0.1-100 (default 5)
bin_values	--	Dictionary of bin_values for converstion of npz to RR-format, see source code (default {})

Benchmark

The benchmark is built on the Pconsc3 dataset and is available for download here. The benchmark contains the input alignment, initial fasta file, predicted distance (both as *.rr and *.npz files) together with one model each for base/dist/trRosetta and Tmscore and timings for each of the 210 proteins. Observe that the times for the base and distance models are for a total of 20 models whereas the timing for trRosetta are for 1 model.

j3xugit/pyconsFold