- Home
- Skills
- Research & Analysis
- SAR report: scaffolds, matched pairs and activity cliffs
Works with the AI Agents you already use
SAR report: scaffolds, matched pairs and activity cliffs
Turn a compound + activity CSV into scaffold groups, matched-pair transformations, activity cliffs and next compounds to make.
$12
SAR report: scaffolds, matched pairs and activity cliffs
Example session with this skill installed
Run a SAR analysis on this EGFR series (IC50 in nM) and tell me what drives potency.
- Read your context and instructions
- Compiled the sar report
- Generated the data export
Summary
37 compounds analysed (0 skipped). pActivity from IC50_nM: min 4.00, median 6.45, mean 6.76, max 10.54, SD 1.53.
5 Murcko scaffolds, 126 matched pairs (82 distinct transformations), 14 activity cliffs (Tanimoto >= 0.7, |delta p| >= 1.0).
Most potent: CHEMBL29197 (10.54), CHEMBL420624 (10.00), CHEMBL328216 (9.92), CHEMBL93032 (9.77), CHEMBL52765 (9.10)
Scaffolds
| scaffold | count | mean_p | max_p | best_id |
|---|---|---|---|---|
| c1ccc(Nc2ncnc3ccccc23)cc1 | 22 | 7.26 | 10.54 | CHEMBL29197 |
| c1ccc(CNc2ncnc3ccccc23)cc1 | 12 | 6.13 | 8.00 | CHEMBL325245 |
| c1ccc(Oc2ncnc3ccccc23)cc1 | 1 | 6.12 | 6.12 | CHEMBL117710 |
| c1ccc(CCNc2ncnc3ccccc23)cc1 | 1 | 5.39 | 5.39 | CHEMBL117804 |
| c1ccc(Nc2ccnc3ccccc23)cc1 | 1 | 5.26 | 5.26 | CHEMBL37346 |
Read: the direct 4-anilinoquinazoline link is worth about 1.1 log over the benzylamine (CH2) version and 1.9 over the homologated CCN. Swapping the quinazoline N3 for CH (last row) costs ~2 log.
Top MMP transformations
| transform | count | mean delta_p | min | max |
|---|---|---|---|---|
| O=N+* >> N* | 3 | +2.32 | -0.09 | +4.00 |
| c1ccc(C*)cc1 >> Brc1cccc(*)c1 | 8 | +1.38 | -0.24 | +3.25 |
| O=N+* >> H | 5 | +1.38 | +1.12 | +1.57 |
| O* >> CO* | 2 | +1.26 | +0.75 | +1.77 |
| CO* >> N* | 5 | +0.85 | -0.85 | +2.64 |
Read: reducing the nitro to an amine is the single biggest consistent gain. The meta-halogen aniline (Br) beats the benzyl across 8 pairs, which is the classic EGFR meta-halide preference.
Activity cliffs (top by SALI)
| id_low | id_high | tanimoto | p_low | p_high | delta_p | sali |
|---|---|---|---|---|---|---|
| CHEMBL289787 | CHEMBL29197 | 0.711 | 6.01 | 10.54 | 4.53 | 15.7 |
| CHEMBL52197 | CHEMBL29197 | 0.717 | 6.46 | 10.54 | 4.08 | 14.4 |
| CHEMBL333851 | CHEMBL115745 | 0.907 | 6.23 | 7.25 | 1.02 | 11.0 |
Next compounds to make
- Apply the nitro -> amine reduction to the best unreduced quinazoline in the series: it is the only transformation with a mean above +2 log.
- Put the meta-Br aniline on the direct N-linked scaffold, combining the two independent gains.
- Resynthesise and retest CHEMBL289787 before trusting the 4.5-log cliff: a jump that large on 0.71 similarity is as likely to be an assay artefact as real SAR.
Files: sar_report.md, scaffolds.csv, mmp_transforms.csv, cliffs.csv (+ 2 more).
Real output: 37 EGFR quinazolines from ChEMBL document CHEMBL1128536.
scaffolds.csv
CSV · data export
Example file from a real run - the skill writes it into your workspace.
Connects securely to your tools. The creator never sees your data.
About this skill
Give your agent a CSV of compounds with activities and get a medicinal-chemistry SAR report back.
What it does
- Normalise: converts IC50/Ki in nM to pActivity (9 - log10 nM), standardizes structures and averages replicate measurements of the same compound.
- Scaffolds: groups by Murcko scaffold with count, mean and max pActivity, and the best compound in each group.
- Matched pairs: finds matched molecular pairs with RDKit rdMMPA single cuts plus H->X sites, keeps small changes only, and ranks transformations by mean delta pActivity with counts and spread.
- Activity cliffs: pairs with Morgan2 Tanimoto >= 0.7 and |delta p| >= 1.0, ranked by SALI.
- Output: a Markdown report, six CSVs and an optional cliff plot.
Why it's trustworthy
All the chemistry and statistics run in a bundled, self-tested script (scripts/sar.py --selftest); the model never estimates a scaffold or a delta. Transformations seen only once are labelled as such, so single observations are not presented as trends.
The agent closes every report with 2-3 concrete compounds to make next, each tied to a transformation in your own data.
Requirements
uv. RDKit and matplotlib install automatically on first run. Input is any CSV with a SMILES column, an id column and an activity column.
- Statistics on your data only: transformations seen once are anecdotes, not trends, and the report labels them as such.
- Treats every activity value as comparable, so mixing assays or cell lines will mislead. Filter before running.
- Replicates of the same standardized structure are averaged.
- 2D only: no 3D alignment, docking or free-energy estimates.
- Matched pairs come from single cuts, so double-cut (linker) changes are not enumerated.
- Needs uv and one-time internet access to install RDKit.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 1 month ago
Requires
Permissions
Shell: runs the bundled scripts/sar.py with uv. Network: only uv downloading RDKit and matplotlib on first run. Read/Write: reads your activity CSV and writes the report, CSVs and optional plot into the output folder you choose.