← Về thư mục
📄 / / profiles / r-and-d / raw_evidence / OpenAI4S / skills / bioskills / bio-chemoinformatics-pharmacophore-modeling / references / usage-guide.md

Pharmacophore Modeling Usage Guide

Overview

Build 3D pharmacophore models from ligands or protein pockets. apo2ph4 derives LigandScout PML models from an apo pocket; PLIP can identify interactions in a co-crystal. Apply a pharmacophore only after representing it in the format required by Pharmer or Pharmit. PharmacoForge can generate candidate pharmacophores conditioned on a protein pocket, followed by compound-library retrieval.

Prerequisites

pip install rdkit
# For PLIP interaction analysis (receptor-based):
pip install plip

Quick Start

Tell the AI agent what to do: - "Derive a pharmacophore from co-crystal of receptor + ligand" - "Build a ligand-based pharmacophore from 5 active compounds" - "Search library for compounds matching pharmacophore query" - "Identify pharmacophore-equivalent bioisosteres"

Example Prompts

Receptor-based pharmacophore

"From co-crystal complex.pdb, identify the bound ligand and compute pharmacophore features (donor, acceptor, hydrophobe, aromatic) at contact points. Use PLIP for interaction analysis."

Ligand-based common pharmacophore

"Given 5 active compounds (SMILES list), align and extract common 3D pharmacophore features. Use RDKit Pharm3D framework."

Pharmacophore screening

"Configure this four-feature pharmacophore in Pharmit and screen the selected library. Report compounds matching at least three features within 1.5 A tolerance and record the exact query representation."

Bioisostere expansion

"Replace each pharmacophore feature with bioisosteric alternatives (carboxylate -> tetrazole, amine -> guanidine). Search for matches."

What the Agent Will Do

  1. Identify pharmacophore features (donor, acceptor, hydrophobe, aromatic) from input.
  2. For a co-crystal: use PLIP or PoseView to map ligand-residue contacts; for an apo pocket, use the documented apo2ph4 workflow and retain its PML output.
  3. For ligand-based: align defensible active conformers, derive conserved feature correspondences and distance bounds with a documented workflow, then use RDKit or another search engine to apply the resulting model. RDKit EmbedPharmacophore embeds against an existing model; it does not derive the consensus.
  4. Set geometric tolerances from aligned-feature variability, coordinate uncertainty, and retrospective validation.
  5. Convert the model to the search engine's documented query format, then search by feature-distance constraints.
  6. Output: ranked hits + retrospective enrichment.

Tips