OpenViSTAI · Agents · Blockchain · Crypto
Submit a project
Menu
AI Models

Deepinteract

DeepInteract is an open-source Python geometric deep-learning framework that uses Geometric Transformers to predict protein interface contacts. Its repository provides scripts for inference and model training, plus data-handling code for protein interaction datasets.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Deepinteract — OpenViST project profile in AI Models.
OpenViST project illustration. See the linked sources for product details.
Understand the project

Practical overview

When it fits

Use DeepInteract when a research workflow needs geometric deep-learning predictions of contacts between two protein chains and can provide the required checkpoints and HH-suite3-compatible sequence database.

Who it is for

  • Bioinformatics researchers studying protein-protein interaction interfaces
  • Machine-learning practitioners developing or evaluating protein contact-prediction workflows

Documented uses

  • Predict a contact probability map from the PDB files of two protein chains.
  • Train, fine-tune or test models using the documented CASP-CAPRI, DB5-Plus or DIPS-Plus targets.
Before you choose

Requirements & limits

The repository is public and identifies GPL-3.0 as its license.

  • Inference requires two chain-specific PDB files, a trained checkpoint and an HH-suite3-compatible database.
  • Database choices have substantially different stated storage requirements: about 17 GB for Small BFD, 86 GB for Uniclust30 and 1.7 TB for BFD, before extraction.
  • The documented traditional installation is for Linux-based operating systems and includes Conda setup and PSAIA dependencies.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Prepare a Small BFD sequence database

Download and unpack the Small BFD database for use with DeepInteract's HH-suite3 database argument.

Before you start

  • aria2c
  • gunzip
  • Approximately 17 GB of storage when the database is unextracted

Steps

  1. Create a local database directory, such as ~/Data/Databases/small_bfd.

  2. Use aria2c to download the Small BFD file from the documented Google Cloud Storage URL into that directory.

  3. Unpack the downloaded .gz file with gunzip.

  4. Use the resulting FASTA file path as the value for --hhsuite_db when following the prediction instructions.

Expected outcome

The Small BFD FASTA file is available at the path specified for DeepInteract's --hhsuite_db argument.

Guide limitations

  • This prepares the database only; it does not install DeepInteract or run a prediction.
  • The README lists BFD and Uniclust30 as alternative compatible databases.
Documentation 1 ↗
Source-based practical guide

Run documented contact-map inference

Generate a predicted contact probability map for two protein chains using a downloaded checkpoint.

Before you start

  • A Linux-based DeepInteract installation and its configured Conda environment
  • Left- and right-chain PDB files
  • A downloaded DeepInteract checkpoint
  • One HH-suite3-compatible sequence database and its local path
  • The repository's documented dependencies, including PSAIA setup

Steps

  1. Download the documented trained checkpoint files into project/checkpoints.

  2. Change to the project directory.

  3. Run lit_model_predict.py with the left and right PDB file paths, checkpoint directory and checkpoint name, and the --hhsuite_db database path.

  4. Find the predicted contact probability map saved as a NumPy array in the specified input directory.

Expected outcome

A predicted protein interface contact probability map, saved as a NumPy array file.

Guide limitations

  • The database download and installation steps must be completed separately.
  • The README notes that python3 lit_model_predict.py --help lists available command-line arguments.
  • The instructions describe a particular checkpoint and sample PDB paths; they do not establish results for other inputs.
Documentation 1 ↗
COMPARE OPTIONS

Compare options for Model inference and serving

These projects document a common task and share a project role. This is a source-based comparison, not an execution benchmark or a guarantee of interchangeability.

ProjectDocumented purposePlatformAccess and requirements
DeepinteractPredict a contact probability map from the PDB files of two protein chains.The repository is public and identifies GPL-3.0 as its license.
Gemma3Moe BuilderConvert a dense Gemma-3-270m checkpoint into a model with a router and configurable expert count and top-k selection.CLI, Linux, macOS, Self-hostedThe repository documents local installation as a Python package and requires Python 3.10 or later. Training and routing tools require the optional `datasets` dependency.
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
18 October 2026
Original website title
BioinfoMachineLearning/DeepInteract · Official homepage title · retained in its source language
How we review sourcesSuggest a correction