Recommended Workflow for Preparing Ligand Libraries with Structural Issues

[zhanglzu]

Hi Uni-Dock2 team,

First, thank you for developing this excellent tool.

I am currently working on a virtual screening project using Uni-Dock2 and have run into a challenge with ligand preparation. My starting point is a large library of ligands in PDBQT format, and my goal is to convert them to SDF for processing. However, I've encountered two main issues that filter out a large portion of my dataset:

1. Poor Initial Geometry in PDBQT Files:
   A significant number of the PDBQT files appear to have unphysical geometries. When I process them, many are rejected due to atoms being too close. For example, I've found ligands with N-H bond distances as short as 0.708 Å, which is chemically unrealistic.

2. Failure to Rebuild from SMILES:
   To solve the geometry issue, I tried a more robust approach: extracting the SMILES string from the PDBQT remarks and rebuilding the 3D structure from scratch using RDKit. While this fixes the simple geometry clashes, this method fails for another large subset of my molecules. RDKit's EmbedMolecule function often fails for molecules with complex or highly strained ring systems.

Given these challenges, I would like to ask for your guidance. Could you please recommend a robust software tool or a standard workflow for cleaning and preparing a large, and potentially "dirty," ligand library for use with Uni-Dock2?

My dataset is downloading from https://virtual-flow.org/real-library，and，unfortunately composed mostly of molecules that fall into one of these two problematic categories, so simply discarding them is not a viable option.

Thank you for your time and any advice you can provide.

[Hong-Rui]

For the ligand preparation issue, we recommended you use our commercial drug design platform, Hermite.
https://www.dp.tech/product/hermite

We start the ligand preparation from SMILES, and we have a refined conformation generation algorithm for complex ring systems.
Hope this helps well.