Disobind is a deep learning method for predicting inter-protein contact maps and interface residues for an IDR and its binding partner from their sequences.
- Kartik Majila, Varun Ullanat, Shruthi Viswanath. Disobind: A sequence-based, partner-dependent contact map and interface residue predictor for intrinsically disordered regions. (2026) Cell Systems).
- Data is deposited in Zenodo
A Google Colab notebook for running Disobind+AF2 is available here. It makes use of the ColabFold implementation of AF2.
- See
requirements.txtfor Python dependencies.
-
Install Conda
If not already installed, install Conda as specified here: https://docs.conda.io/projects/conda/en/latest/index.html. -
Clone the repository
git clone https://github.com/isblab/disobind.git
- Set up the repository
Run the following commands in order:
cd disobind/
chmod +x install.sh
./install.sh
For using GPUs, ensure CUDA-toolkit (version 11.8) and the NVIDIA drivers are installed on the system.
- Disobind can only be used for binary complexes (AB). However, for non-binary complexes (ABC) the user can convert them into binary pairs (AB, BC, AC) to run Disobind.
- The input protein pair is assumed to be interacting and Disobind predicts where they interact (contact maps and interface residues).
- Protein 1 must be an IDR whereas Protein 2 may or may not be an IDR.
The input is a CSV file.
Each row corresponds to one sequence fragment pair for which the Disobind prediction is required.
Each row contains the UniProt ID, start, and end UniProt residue positions for each of the two protein sequence fragments.
To run a Disobind prediction only, provide the input as:
UniProt_ID1, start1, end1, UniProt_ID2, start2, end2.
To run a Disobind+AF2 prediction, provide the input as:
UniProt_ID1, start1, end1, UniProt_ID2, start2, end2, AF2_struct_file_path, AF2_pae_file_path, chain1, chain2, offset1, offset2.
Chain1, Chain2 represent the Chain IDs that correspond to the protein1/2 sequence fragment.
Offset1, Offset2 are integer values that indicate the difference in the residue positions between the AF2 structure and UniProt position.
Set the offsets to 0 if the AF2 structure corresponds to the full UniProt sequence or just the sequence fragment.
As an example see example/test.csv.
Run the following command to use Disobind for the example case with default settings:
python run_disobind.py -f ./example/test.csv
By default, Disobind provides interface predictions at a coarse-grained (CG) resolution 1.
| Flags | Description |
|---|---|
| -f | path to the input csv file. |
| -c | no. of cores to be used for downloading the UniProt sequences (default = 2). |
| -o | output directory name (default: output). |
| -d | device to be used - cpu/cuda (default: cpu). |
| -cm | whether to predict inter-protein contact maps (default: False). By default, only interface residues are predicted. |
| -cg | coarse-grained resolution - 0, 1, 5, 10 (default: 1). If 0, predictions at all resolutions (1,5 and 10) are provided. |
This script outputs the following files:
-
A CSV output file for all predictions for all input sequence fragment pairs. See the Colab notebook for description of the output CSV.
-
Predictions.npy: contains predictions for all input sequence fragment pairs in a nested dictionary.
The output CSV file contains the following four columns:
| Protein1 | Residue1 | Protein2 | Residue2 |
|---|---|---|---|
| X1 | 10 | X2 | 40 |
| X1 | 14 | X2 | 44 |
| X1 | 125 | X2 | 80 |
For contact map prediction, this must be interpreted as, residue 10 in protein X1 interacts with residue 40 in protein X2 and so on.
For interface residue prediction, this must be interpreted as, residues 10, 14, and 125 in protein X1 may interact with one or more of the residues 40, 44, and 80 in protein X2.
Follow the steps as specified in dataset.
Follow the steps as specified in src.
Follow the steps as specified in analysis.
Author(s): Kartik Majila, Varun Ullanat, Shruthi Viswanath
Date: Jan 10, 2026
License: GPL v3 This work is licensed under the terms of the GNU General Public License, Version 3, as published by the Free Software Foundation on 29 June 2007.
Testable: Yes
Parallelizeable: Yes
Publications: Majila K., Ullanat V., Viswanath S. Disobind: A sequence-based, partner-dependent contact map and interface residue predictor for intrinsically disordered regions. CellSystems (2026), DOI.