sar-analysis

SAR Analysis

Metadata

Short Description: Comprehensive guide for performing Structure-Activity Relationship (SAR) analysis using RDKit.

Authors: Ohagent Team

Version: 1.0

Last Updated: December 2025

License: CC BY 4.0

Commercial Use: ✅ Allowed

Overview

Structure-Activity Relationship (SAR) analysis is a core medicinal-chemistry workflow that relates systematic structural variations of a chemical series to changes in biological activity. The goal is to (1) identify a common scaffold (Maximum Common Substructure, MCS) shared by a series of analogues, (2) decompose each molecule into the scaffold plus its R-group substituents, (3) align all molecules so substituents at equivalent positions are visually comparable, and (4) connect substituent variation to potency to derive testable design hypotheses.

This guide formalizes a reproducible RDKit-based SAR workflow that produces an interactive HTML report (compound table with aligned core/R-groups and an activity heatmap) and a written SAR narrative that explicitly contrasts substituents at the same R-position. It is intended for use on activity tables containing SMILES, a compound identifier, and a numeric potency value (IC50, Ki, EC50, %inhibition, etc.).

Key Concepts

Maximum Common Substructure (MCS)

MCS is the largest connected substructure shared by all (or a configurable threshold of) molecules in a set. RDKit's rdFMCS.FindMCS searches for this scaffold under tunable atom/bond comparison rules. For SAR, MCS provides the anchor template against which every analogue is decomposed and aligned. A threshold=0.8 allows MCS to be defined when only 80% of molecules contain the candidate substructure, which is more robust to outliers than threshold=1.0. ringMatchesRingOnly=True and completeRingsOnly=True prevent partial-ring fragments that look chemically meaningless.

R-Group Decomposition

R-group decomposition (rdRGroupDecomposition.RGroupDecompose) maps each molecule onto the MCS core and assigns the non-core fragments to enumerated R-positions (R1, R2, …). The output is a per-molecule dictionary {Core, R1, R2, …}. Constant R-positions (where every molecule carries the same fragment) are uninformative for SAR and should be pruned from the report so attention focuses on the variable positions that actually drive activity.

Substructure Alignment for Comparable 2D Depiction

For SAR visualization to be interpretable, the core and each R-group must be drawn in the same orientation as the parent molecule. The recommended pattern uses three fall-back strategies in order: (1) a direct GetSubstructMatch, (2) a re-match after AdjustQueryProperties(makeDummiesQueries=True) so R-group dummy atoms are treated as queries, and (3) a final attempt with useChirality=False. Once a match is found, atom coordinates are copied from the parent conformer onto the fragment. Without this, R-group cells are drawn in arbitrary canonical orientations and visual SAR is essentially impossible to read.

Activity Heatmap and Comparative Analysis

A logarithmic-scale color gradient (green = high potency / low IC50, red = low potency / high IC50) on the activity column lets a reader spot trends across the series at a glance. The accompanying narrative must justify every claim about a substituent's effect by explicit pairwise contrast at the same R-position — the unit of SAR evidence is "compound A (R1=X, IC50=…) vs compound B (R1=Y, IC50=…)", never an unsupported generalization.

Decision Framework

SAR analysis pipeline
└── Have SMILES + activity for >= 4 analogues?
    ├── No  -> Insufficient data; collect more analogues first
    └── Yes -> Run rdFMCS.FindMCS(threshold=0.8, ringMatchesRingOnly=True)
        ├── MCS too small (<5 atoms) -> Series is too diverse;
        │                               cluster first, then run SAR per cluster
        └── MCS reasonable -> RGroupDecompose
            └── For each fragment alignment to parent:
                ├── Strategy 1: GetSubstructMatch(direct)  -- works for canonical cases
                │   └── No match -> Strategy 2
                ├── Strategy 2: AdjustQueryProperties(makeDummiesQueries=True)
                │               -- handles dummy R-group atoms
                │   └── No match -> Strategy 3
                ├── Strategy 3: GetSubstructMatch(useChirality=False)
                │               -- handles stereochemistry mismatches
                │   └── No match -> Compute2DCoords as fallback (lose alignment)
                └── Drop constant R-positions, build HTML, draw with DrawMoleculeACS1996

| Situation | Recommended choice | Rationale | |-----------|--------------------|-----------| | Standard congeneric series with a clear scaffold | MCS threshold=0.8, ringMatchesRingOnly=True, completeRingsOnly=True | Tolerates a small minority of outliers while keeping rings intact | | Highly diverse set (e.g., HTS hit list) | Cluster (Tanimoto/Murcko) first, then SAR per cluster | A single MCS will be too small to be useful across diverse chemotypes | | Stereoisomers in the series | Try Strategy 1 first; fall back to Strategy 3 (useChirality=False) | Chirality differences should not break depiction alignment | | Analogues with R-group attachment dummies in queries | Strategy 2 with AdjustQueryProperties(makeDummiesQueries=True) | Dummy atoms are treated as queries so they match real heavy atoms | | One R-position constant across all analogues | Drop from report and from core depiction | Constant positions are uninformative and clutter the table | | Activity spans many orders of magnitude | Color heatmap on log10(activity) | Linear scale collapses the dynamic range visually | | Drawing for publication or report | DrawMoleculeACS1996 via MolDraw2DSVG | ACS1996 is the de facto standard for medicinal chemistry figures |

Best Practices

1. Inspect the dataframe before assuming column names. Real-world activity tables vary; auto-detect the SMILES, activity, and ID columns from df.head() rather than hard-coding names. This avoids silent failures on user data.
2. Add explicit hydrogens before MCS. Chem.AddHs lets MCS reason correctly about heavy-atom valence and ring closures; without it, otherwise-identical scaffolds can be missed.
3. Prune constant R-positions. Any R-position whose fragment is identical across every analogue contributes no SAR information; remove that column from the table and remove the constant attachment point from the core depiction so the variable positions stand out.
4. Always align fragments to the parent molecule, not the other way around. Copy coordinates from the parent onto each fragment via the matched atom map. Drawing the parent canonically and then re-drawing fragments from scratch loses comparability between rows.
5. Use a log-scale activity heatmap. Potency typically spans 2–4 orders of magnitude; a linear color scale collapses the interesting low-IC50 region. Map green to low IC50 (high potency) and red to high IC50 (low potency).
6. Justify every SAR claim with a pairwise contrast at the same R-position. Statements like "small electron-withdrawing groups improve activity at R1" must be backed by a direct comparison such as "compound 7 (R1=F, IC50=0.5 µM) vs compound 1 (R1=Me, IC50=5.2 µM)". Unsupported generalizations are not acceptable evidence.
7. Test 3-4 analogues per design hypothesis. A single substitution change can be confounded by experimental noise; multiple analogues at the same position give a defensible trend.
8. Render with DrawMoleculeACS1996. ACS1996 styling produces consistent bond lengths, atom labels, and font choices that match medicinal-chemistry publication norms; avoid mixing styles within a single report.

Common Pitfalls

• Pitfall: Hard-coding column names like "SMILES" or "IC50". Different vendors and ELNs export different headers; the script br