I am trying to replicate a successful KNIME workflow that uses Flare and other nodes to trim down chemical libraries and dock them.
While I do appreciate Ligandscout does provide a Vina node, I am trying to avoid subscriptions/licenses to run opensource.
At the moment the issue seems to be with the pdbqt format that Vina needs.
The workflow generates sdf and pdb that I then convert to pdbqt via the OpenBabel node.
So far so good.
The challenge is to perform the conversion of multiple molecules contained in the same sd file to then generate x number of pdbqt files for Vina to dock.
I believe Vina is not really geared to virtual screen mode but it can be forced to dock each one in a batch mode but have no experience of this.
While I could try to do this via the CLI, the aim is to deploy it in a Knime workflow.
Would you maybe have a short workflow with a few molecules that demonstrates what you have done so far (including a Conda Environment Propagation node with your env)?
It requires a Conda environment with several packages installed, and uses Python Scripting nodes for the docking step with Gnina, as well as a few post-processing tools, including PoseBusters, clustering, and an RMSD comparison between the poses and the reference ligand.
All tools are open-source, so no commercial licenses needed. Let me know if you need any help, I have not come around to documenting everything.
thank you Evert, coming to the rescue once more!
Great to see Gnina being incorporated in a Knime workflow - will definitely give it a go.
I am also planning to use the Docker image to minimise cross-contamination and keep IT happy (DiffDock in Docker worked a charm).
I am assuming Gnina can also use GPUs like Diffdock - any experience?
Gnina can run without GPU but is much faster with GPU. It can run on multiple CPUs in parallel though. I am also working on a component for calculating protein-ligand interaction fingerprints for the docking poses, and comparing them with the reference ligand, but coding this has proven more difficult for the various AIs I have tried than anticipated. Stay tuned.
I uploaded an updated version that has a component for calculating and comparing protein-ligand interaction fingerprints (PLIFs) between the reference ligand and the docking poses.
Pareto scoring on a combination of parameters is then used to prioritize the ligands.
With PyMol installed in the Conda environment, running either the FPocket or Gnina Docking component should result in PyMol being launched and the results being visualized.
