Q: Can DiffDock predict the binding affinity of a ligand to a protein?

No, DiffDock only predicts the 3D binding structure and outputs a confidence score, not binding affinity. The FAQ recommends combining it with other tools like GNINA or MM/GBSA for affinity estimates.

Q: Is DiffDock suitable for docking peptides or large biomolecules?

It's designed for small molecules only; for peptides or larger biomolecules, the FAQ suggests alternatives like DiffDock-PP for protein-protein or RoseTTAFold2NA for nucleic acid interactions, as performance may be unreliable.

Q: How to use DiffDock with a protein sequence instead of a PDB file?

Provide the protein sequence via '--protein_sequence' flag, and DiffDock will fold it using ESMFold internally. This allows docking without pre-existing protein structures, useful for novel targets.

Question 1

How does DiffDock compare to traditional docking software like AutoDock Vina?

Accepted Answer

DiffDock uses a diffusion model for higher accuracy and provides confidence scores, while traditional tools like Vina rely on scoring functions and search algorithms, often with lower accuracy but faster runtime. DiffDock is state-of-the-art but may be slower without GPU.

Question 2

How to install and run DiffDock locally on a Linux system?

Accepted Answer

Clone the repo, set up a conda environment using environment.yml, activate it, and run inference via CLI with commands like 'python -m inference'. For GPU support, ensure CUDA is installed, and use Docker if preferred for containerized deployment.

Question 3

What does the DiffDock confidence score mean and how should I interpret it?

Accepted Answer

The confidence score indicates the model's certainty in the predicted pose, with guidelines in the FAQ: >0 for high confidence, -1.5 to 0 for moderate, and <-1.5 for low, but it varies with ligand size and protein conformation.

Question 4

Can DiffDock predict the binding affinity of a ligand to a protein?

Accepted Answer

No, DiffDock only predicts the 3D binding structure and outputs a confidence score, not binding affinity. The FAQ recommends combining it with other tools like GNINA or MM/GBSA for affinity estimates.

Question 5

Is DiffDock suitable for docking peptides or large biomolecules?

Accepted Answer

It's designed for small molecules only; for peptides or larger biomolecules, the FAQ suggests alternatives like DiffDock-PP for protein-protein or RoseTTAFold2NA for nucleic acid interactions, as performance may be unreliable.

Question 6

How to use DiffDock with a protein sequence instead of a PDB file?

Accepted Answer

Provide the protein sequence via '--protein_sequence' flag, and DiffDock will fold it using ESMFold internally. This allows docking without pre-existing protein structures, useful for novel targets.

DiffDock

What is DiffDock?

Overview

Use Cases

Best For

Found a gem we're missing?

Not Ideal For

Pros & Cons

Pros

Cons

Frequently Asked Questions