Use ColabDock Online

ColabDock: Restraint-Guided Protein Complex Docking with AlphaFold2

ColabDock is a restrained protein complex prediction framework that inverts the AlphaFold2 structure prediction model for protein-protein docking with sparse experimental restraints. It was developed by Shihao Feng, Zhenyu Chen, Chengwei Zhang, Yuhao Xie, Sergey Ovchinnikov, Yiqin Gao and Sirui Liu, with affiliations including Changping Laboratory, Peking University and Harvard University, and is built on ColabDesign.

Experiments such as NMR chemical shift perturbation, covalent labeling, cross-linking mass spectrometry and deep mutational scanning give partial information about where two proteins touch, but rarely enough to define a full complex. Traditional docking programs such as HADDOCK and ClusPro use such restraints, yet the authors report that ColabDock outperforms them across several restraint types.

ColabDock takes the individual chain structures as templates and combines them with whatever sparse restraints you have, so the predicted complex stays consistent with both the experimental data and AlphaFold2 knowledge of protein structure.

How ColabDock Works: Generation, Prediction and Ranking

The method has two stages followed by ranking.

1. Generation stage: the input sequence is optimized in a continuous logit space to fit the chain templates and restraints, using four losses: monomer distogram, restraint, pLDDT and interface predicted aligned error (ipAE).

2. Prediction stage: AlphaFold2 predicts the final complex from the chain templates and the generated complex.

3. Ranking: multiple rounds produce diverse structures, and RankingSVM models select the top candidates using features such as ipTM, contact number, contact pLDDT and the number of satisfied restraints.

Two restraint formats

A 1v1 restraint links one specific residue pair, as in cross-linking data. An MvN restraint says a residue in one chain contacts any of a set of residues in the other chain, which suits NMR, covalent labeling and DMS data. Defaults are 15 rounds for 1v1 and 10 for MvN restraints, with 50 steps per round.

ColabDock Benchmark Results Versus HADDOCK and ClusPro

On the benchmark set with 1v1 restraints, ColabDock reached a mean DockQ of 0.477, compared with 0.287 for HADDOCK and 0.190 for ClusPro. With only two restraints, it produced acceptable or better predictions for 23 of 37 complexes, versus 14 for HADDOCK and 7 for ClusPro.

With NMR restraints, ColabDock reached a DockQ of 0.603 on EIN and 0.936 on E2A, against 0.385 and 0.411 for HADDOCK. It also tolerated loose restraints in the 8-20 Å range, reaching a mean DockQ of 0.344 across 176 samples. For antibody-antigen complexes with simulated DMS-style restraints, ColabDock gave acceptable or better top1 predictions for 43 cases, against 32 for HADDOCK and 28 for ClusPro.

What is Tamarind Bio?

Tamarind Bio is a no-code bioinformatics platform built to give life scientists and researchers access to powerful computational tools. Many cutting-edge machine learning models are hard to deploy and use. Tamarind provides an intuitive, web-based environment that removes the complexity of high-performance computing, software dependencies and command-line interfaces.

The platform is designed for biologists, chemists and other researchers who may not have a background in programming or cloud infrastructure but want to run models on their own data. Key features include:

• A user-friendly graphical interface for setting up and launching experiments

• A robust API for integration into existing research pipelines

• An automated system for managing and scaling computational resources

Tamarind treats information and data security as a top priority, as detailed in its Trust Center and Terms of Service.

Accelerating Discovery with ColabDock on Tamarind Bio

• NMR-guided complexes: turn chemical shift perturbation data into a structural model of the interface.

• Covalent labeling and cross-linking: combine footprinting or distance data with AlphaFold2 knowledge to model assemblies.

• Antibody-antigen interfaces: use DMS-style information about which residues matter to improve interface prediction.

• Loose or partial restraints: get usable models even when only a few, imprecise contacts are known.

How to Use ColabDock on Tamarind Bio

1. Log in: sign in to tamarind.bio and open the ColabDock tool.

2. Provide chain structures: upload the structure of each protein chain, which ColabDock uses as templates.

3. Define restraints: enter residue-pair (1v1) or residue-to-set (MvN) restraints from your experiment, if exposed.

4. Set parameters: adjust options such as number of rounds or restraint distance only if exposed; the paper defaults are 15 rounds for 1v1 and 10 for MvN.

5. Run: submit the job and wait for the ranked models.

6. Download outputs: retrieve the predicted complex structures.

7. Interpret: check how many restraints each model satisfies, along with confidence scores such as ipTM.

8. Use downstream: take top models into interface analysis, mutagenesis planning or design.

Things to Keep in Mind

• Restraints are limited to residue pairs under 22 Å, set by the AlphaFold2 distogram upper bound.

• Complexes must be under 1200 residues on an NVIDIA A100 GPU, and large complexes can run for a long time.

• Antibody-antigen accuracy remains modest in absolute terms (mean DockQ 0.223).

• The reported version is a preprint that has not been peer reviewed.

Source: Feng S, Chen Z, Zhang C, Xie Y, Ovchinnikov S, Gao Y, Liu S. ColabDock: inverting AlphaFold structure prediction model for protein-protein docking with experimental restraints. bioRxiv preprint, 2023. DOI: 10.1101/2023.07.04.547599. Code (stated by the authors): github.com/JeffSHF/ColabDock.