Virtual Screening Workflow

Drug discovery increasingly benefits from access to vast chemical spaces, driven by generative AI and, more importantly, the systematic enumeration of synthetically feasible molecules. Meanwhile, advances in AI-based protein structure prediction and co-folding have enabled docking-based virtual screening against virtually any biological target.

This tutorial provides a practical, step-by-step workflow for performing high-throughput virtual screening, including:

• 🧹 Library filtering
• 🌀 3D conformer generation with pH-dependent protonation state assignment
• 🧬 Protein preparation
• ⚡ High-throughput molecular docking
• 📊 Post-docking analysis and hit prioritization

---

📚 Example Library

The toy library contains 100 compounds, including two compounds previously identified through virtual screening against the kinase domain of EPHB4. One of these compounds subsequently contributed to the discovery of a clinical candidate for cancer treatment (The science and art of structure-based virtual screening).

This example demonstrates how a compound library can be processed, docked, and analyzed using the LeDock_NOVA virtual screening pipeline.

---

1. 🧹 Library Filtering

Although higher-quality hits can often be identified by screening larger libraries, docking the entire chemical library is not always necessary or computationally efficient.

It is therefore recommended to prefilter compound libraries based on desired physicochemical properties and/or pharmacophoric features before docking.

Command

cd library

../tools/queryDB -filter ../param/library_filter.param smi lib.smi lib_f.smi

The library_filter.param file contains general-purpose filtering criteria, including:

• Removal of problematic substructures (e.g., PAINS)
• Desired (privileged) substructures
• Lead-like physicochemical property filters

In a real drug discovery project, these parameters should be carefully optimized according to the specific target, therapeutic area, and project objectives.

---

2. 🌀 3D Conformer Generation

Generate 3D conformers with pH-dependent protonation state assignment:

Command

../tools/leconf -i lib_f.smi -o lib_f.sdf --n 1 --t

For each molecule, leconf:

• Samples multiple conformations and selects the lowest-energy conformer (--n 1)
• Assigns protonation states at physiological pH (7.4)
• Selects the lowest-energy ring tautomer (--t)
• Generates energetically favorable chair conformations for six-membered saturated rings
• Preferentially selects equatorial ring substitutions while considering axial orientations caused by allylic strain (The role of allylic strain for conformational control in medicinal chemistry)

The generated conformers are stored in lib_f.sdf for docking.

---

3. 🧬 Protein Preparation

Prepare the protein structure for docking:

Command

cd ../protein
../tools/lepro_nova 2VWX.pdb

lepro_nova performs lightweight protein preprocessing, primarily by adding hydrogen atoms. It also generates the docking configuration file:

dock.in

By default, 20 docking poses are generated per ligand.

In some benchmarking studies (e.g., DUD-based evaluations), only one pose is generated to reduce computational cost and enable large-scale performance comparisons. However, this is generally not recommended for real-world virtual screening, where at least 5–10 poses per ligand are typically required to achieve reasonable sampling convergence.

---

4. ⚡ Docking of the Compound Library

Prepare the ligand list:

Command

ls ../library/lib_f.sdf > ligands

Run parallel docking:

Command

OMP_NUM_THREADS=4 ../tools/ledock_nova dock.in

The number of OpenMP threads can be adjusted according to available CPU resources.

---

5. 📊 Post-docking Analysis

Perform pose filtering and hit prioritization:

Command

../tools/lepose lib_f_dock.sdf ../param/post_docking.param

lepose filters docking poses according to:

• Scoring components
• Key protein–ligand interactions
• Allowed low-energy torsion angles

For kinase targets, example selection criteria may include:

• ✅ Hydrogen bonding interactions with the hinge region
• ✅ Occupancy of the ATP back pocket by aromatic groups

The final output is:

lib_f_dock_pass.sdf

The selected hits can be directly visualized using the ICM Browser (ICM Browser) or other molecular visualization software.

---

🚀 Summary

This workflow provides an end-to-end pipeline for high-throughput virtual screening:

📚 Compound Library
        │
        ▼
🧹 Library Filtering
        │
        ▼
🌀 3D Conformer Generation
        │
        ▼
🧬 Protein Preparation
        │
        ▼
⚡ Molecular Docking
        │
        ▼
📊 Post-docking Analysis
        │
        ▼
🎯 Prioritized Hits

The workflow is designed for rapid screening of large compound libraries while maintaining chemically meaningful filtering and biologically relevant pose selection.