proABC-2

Predicts the antibody residues that will make contact with the antigen and the type of interaction using a Convolutional Neural Network (CNN).

Installation

proABC-2 is available both locally as a python package and as a Docker container. See below instructions for each case.

Docker

The docker image is available on the Github Container Registry and can be pulled using the following command:

docker pull ghcr.io/haddocking/proabc-2:latest

Local

proABC-2 has some third-party dependencies that must be installed before running the software.

proABC-2 is available on PyPI and can be installed using pip using Python3.7:

pip install proabc-2

It also depends on two third-party software, HMMER and IGBLAST, check the third-party section for more information.

Example (Local and Docker)

Set up the data to run the example:

• Create a folder named proabc2-prediction in the root directory.

mkdir proabc2-prediction

• Create a heavy and light fasta file inside proabc2-prediction with the following content:

echo ">APDB_H\nEVQLVESGGGLVQPGGSLRLSCAASGYTFTNYGMNWVRQAPGKGLEWVGWINTYTGEPTYAADFKRRFTFSLDTSKSTAYLQMNSLRAEDTAVYYCAKYPHYYGSSHWYFDVWGQGTLVTVSS" > proabc2-prediction/heavy.fasta

echo ">APDB_L\nDIQMTQSPSSLSASVGDRVTITCSASQDISNYLNWYQQKPGKAPKVLIYFTSSLHSGVPSRFSGSGSGTDFTLTISSLQPEDFATYYCQQYSTVPWTFGQGTKVEIKRTV" > proabc2-prediction/light.fasta

Execution (Docker)

docker run \
  --rm \
  --user $(id -u):$(id -g) \
  -v `pwd`:/data \
  ghcr.io/haddocking/proabc-2:latest \
  proabc2-prediction/ heavy.fasta light.fasta

Execution (Local)

proabc2 proabc2-prediction/ heavy.fasta light.fasta

Results

The output will be in the same folder as the input files, named as heavy-pred.csv and light-pred.csv.

They consist of several columns:

• Chothia: position of the residue according to the Chothia numbering scheme
• Sequence: residue type for each position
• pt: probability of making a general interaction with the antigen
• hb: probability of making a hydrogen bonds with the antigen
• hy: probability of making a hydrophobic interaction with the antigen

$ head proabc2-prediction/*pred.csv
==> proabc2-prediction/heavy-pred.csv <==
,Chothia,Sequence,pt,hb,hy
0,1,E,0.24,0.18,0.24
1,2,V,0.25,0.15,0.25
2,3,Q,0.16,0.16,0.17
3,4,L,0.14,0.14,0.17
4,5,V,0.14,0.15,0.15
5,6,E,0.16,0.16,0.16
6,7,S,0.14,0.16,0.13
7,8,G,0.17,0.13,0.16
8,9,G,0.14,0.14,0.15

==> proabc2-prediction/light-pred.csv <==
,Chothia,Sequence,pt,hb,hy
0,1,D,0.25,0.18,0.2
1,2,I,0.23,0.15,0.2
2,3,Q,0.15,0.16,0.17
3,4,M,0.15,0.14,0.15
4,5,T,0.16,0.15,0.16
5,6,Q,0.15,0.16,0.14
6,7,S,0.15,0.14,0.12
7,8,P,0.15,0.14,0.13
8,9,S,0.14,0.14,0.14

proABC-2 also accepts the DNA sequences of the antibody chains and uses the Biopython Seq module for the translation into protein sequences.

Citation

• F. Ambrosetti, T.H. Olsen, P.P. Olimpieri, B. Jiménez-García, E. Milanetti, P. Marcatilli, A.M.J.J. Bonvin. "proABC-2: PRediction Of AntiBody Contacts v2 and its application to information-driven docking", Bioinformatics, , btaa644,