mmpa

Matched Molecular Pair Analysis (MMPA)

MMPA identifies pairs of molecules (A, B) that differ by a single structural transformation (R₁ → R₂) at one site while sharing an identical molecular context. Aggregating ΔProperty across hundreds of such pairs extracts robust, context-free SAR rules.

When to Use This Skill

• Extract SAR rules from a compound dataset (e.g., which H→F swap improves metabolic stability?)
• Identify activity cliffs (large ΔActivity for small structural change)
• Find bioisosteric replacements supported by experimental data
• Predict property changes for a proposed structural modification
• Generate analogue libraries from a lead compound using proven transforms
• Prioritize which substituent to try next based on MMPA-derived rules

Core Concept

Molecule A:  [Context]-[R₁]   →   transform: R₁ → R₂
Molecule B:  [Context]-[R₂]

ΔProperty = P(B) - P(A)

SAR rule: "R₁ → R₂ causes ΔlogP = +0.45 ± 0.12 (N=18 pairs)"

Fragmentation (single-cut):

• Variable part: the part that changes between A and B
• Context: the shared skeleton (everything outside the cut bond)
• Represented via SMIRKS: [R₁:1]>>[R₂:1] at attachment point

Double-cut: both variable parts and a central linker can vary — rarer, more specific.

Quick Start — mmpdb 4

pip install mmpdb
# or: conda install -c conda-forge mmpdb

# Full pipeline: SMILES file → SAR rules
mmpdb fragment compounds.smi -o fragments.h5
mmpdb index fragments.h5 -o mmpdb.db

# Load experimental properties (CSV with: smiles, id, prop1, prop2)
mmpdb loadprops mmpdb.db properties.csv

# Query: what transforms improve LogD?
mmpdb transform --smiles "c1ccc(cc1)C(=O)O" mmpdb.db \
    --property LogD \
    --min-pairs 3 \
    -o transform_results.csv

# Analyze SAR rules for a property
mmpdb analyze mmpdb.db --property pIC50 -o sar_rules.csv

Quick Start — Programmatic (RDKit + mmpdb Python API)

import mmpdblib
from mmpdblib import do_fragment, do_index

# Fragment a SMILES list programmatically
from mmpdblib.analysis_algorithms import find_mmps

smiles_dict = {
    "mol_A": "c1ccc(CC)cc1",   # ethylbenzene
    "mol_B": "c1ccc(CF)cc1",   # fluoromethylbenzene
    "mol_C": "c1ccc(CCl)cc1",  # chloromethylbenzene
}

# Find matched molecular pairs
pairs = find_mmps(list(smiles_dict.values()), list(smiles_dict.keys()))
# pairs: list of (id1, id2, transform_SMIRKS, context_SMILES)