Loading...
Home About / Validation Documentation Contact Team
Access Results

How can we help you?

Popular: • •

General Information

SEARCH-ML is an automated High-Throughput Virtual Screening (HTVS) pipeline. Unlike traditional docking which simulates physics, we use Machine Learning (XGBoost, Random Forest) trained on massive protein-ligand datasets to predict binding affinity instantly.
Yes, SEARCH-ML is currently free for academic and non-profit research purposes. We believe in open science.
No. SEARCH-ML is entirely cloud-based. All calculations (featurization, prediction) happen on our high-performance cluster (HPC). You can even run it from a mobile phone browser.
Chaurasia, D.K. et al., (2025). "SEARCH-ML: A Stacked Ensemble Framework for High-Throughput Virtual Screening". Frontiers in Molecular Biosciences.

Methodology & Inputs

  • Forward Screening: You have a Target Protein and want to find drugs that fit into it. (Target → Drugs)
  • Reverse Screening: You have a Drug Molecule and want to find which human proteins it binds to (Target Identification). (Drug → Targets)
Input TypeAccepted Formats
Protein Structure.pdb
Small Molecule.sdf, .mol, .smi (SMILES)

The Ligand Code is a unique 3-letter identifier (e.g., ATP, LIG) found inside the PDB file that identifies the small molecule bound to the protein.

To find it: Open your PDB file in a text editor or viewer (PyMOL) and look for HETATM lines.

Binding affinity (ΔG) is usually negative. It represents energy released.

  • More Negative (e.g., -10.0): Stronger binding. The drug fits tightly.
  • Less Negative / Positive (e.g., -2.0 or +5.0): Weak or no binding. The drug falls off.
Raw PDB files often contain water molecules, ions, and crystallization artifacts. Our pipeline removes water and non-essential ions (except those crucial for binding like Zinc/Magnesium) to ensure accurate feature calculation.

Troubleshooting

Cause:

You uploaded a protein-only structure (apo-protein) or the ligand name you provided doesn't match the PDB file content.

Solution:

1. Open your PDB file and ensure HETATM lines exist.
2. Ensure the 3-letter code you typed (e.g., 'LIG') matches exactly what is in the file.

This is usually due to a network interruption or a very large file.

  • Refresh the page and try again.
  • Ensure your PDB file is under 10MB (most proteins are < 1MB).
  • Disable ad-blockers that might block the upload script.
The system is strict about extensions. A PDB file must end in .pdb, not .ent or .txt. Rename your file and try again.
Our ML models are deterministic, so results should be identical. However, if we updated our backend models (e.g., version 1.0 to 1.1), scores might shift slightly. Check the footer for version information.
No. We require 3D coordinate files (SDF, PDB) or SMILES strings. Images (JPG, PNG) do not contain the atomic data required for calculation. Use a tool like ChemDraw or MarvinSketch to convert images to SMILES.

Data & Privacy

Yes. Files are processed in isolated containerized environments. They are never shared publicly and are not used to retrain our public models.
We have a 30-day retention policy. All input files and generated result reports are automatically wiped from our servers 30 days after the job completes.
Currently, manual deletion is not available via the web interface, but you can email support with your Job ID to request an immediate purge.