← Back to README · Documentation index
The demo/ directory contains a full end-to-end run against JAK2 (ChEMBL target CHEMBL2971), exercising curation, tokenization, distribution filtering, OpenEye conformer generation, and the scaffold + split demo plugins.
# MolForge demo: JAK2 (CHEMBL2971) IC50 pipeline
# Full pipeline showcasing curation, tokenization, distribution filtering,
# OpenEye conformer generation, and the scaffold + split demo plugins.
version: "1.0"
steps:
- source
- chembl
- curate
- tokens
- distributions
- confs
- scaffold
- split
source:
backend: sql
version: 36
auto_download: false
chembl:
standard_type: IC50
standard_units: nM
standard_relation: "="
assay_type: B
target_organism: Homo sapiens
assay_format: protein
std_threshold: 0.5
range_threshold: 0.5
curate:
mol_steps: [desalt, removeIsotope, neutralize, sanitize, handleStereo]
smiles_steps: [canonical]
stereo_policy: assign
dropna: true
tokens:
dynamically_update_vocab: true
distributions:
properties: all
global_statistical_threshold: 2.0
curate_tokens: true
filter_unknown_tokens: true
plot_distributions: true
confs:
backend: openeye
max_confs: 100
mode: classic
scaffold:
include_generic: true
split:
test_ratio: 0.1
val_ratio: 0.1
output:
dir: demoThe split step runs last and partitions the final dataset into train, validation, and test sets.
Run it with:
molforge run CHEMBL2971 --config demo/jak2_ic50.yamlSee the CLI reference for the molforge run options and the Python API for the equivalent programmatic entry point.
| Step | In → Out | Notes |
|---|---|---|
source |
— → 56,207 | Raw JAK2 activities fetched from the local ChEMBL 36 database |
chembl |
56,207 → 8,060 | IC50 / binding / protein-format curation, IC50 → pIC50 |
curate |
8,060 → 7,970 | Desalt, neutralize, sanitize, stereo assignment |
tokens |
7,970 → 7,970 | 48-token vocabulary built |
distributions |
7,970 → 5,292 | 2,678 removed (66.4% overall pass rate, ±2σ) |
confs |
5,292 → 5,259 | OpenEye OMEGA, 33 failed, avg ~93 conformers/molecule |
scaffold |
5,259 → 5,259 | 2,018 scaffold clusters |
split |
5,259 → 5,259 | train=4,206, val=527, test=526 |
Total wall-clock time: ~1m 55s.
The pipeline generates a distribution plot for each of the 18 molecular properties, helping identify outliers and validate filtering thresholds:
The split plugin writes a report card summarising split composition, scaffold statistics, and structural separation (the nearest-neighbour ECFP4 Tanimoto distance matrix):
A run writes a self-contained directory under the configured output root. The final curated dataset is CHEMBL2971_ID*.csv; the run's resolved configuration, logs, vocabularies, split report, distribution plots, and conformers accompany it:
demo/CHEMBL2971_ID3B9CE574/
├── CHEMBL2971_ID3B9CE574.csv # Final curated dataset
├── CHEMBL2971_ID3B9CE574.log # Execution log
├── CHEMBL2971_ID3B9CE574_config.json # Resolved pipeline configuration
├── curation_results.json # Distribution curation metadata & statistics
├── vocab.json # SMILES vocabulary
├── vocab_curated.json # Post-curation vocabulary
├── split_report.json # Scaffold-split report card
├── split_report.png # Scaffold-split report figure
├── distributions/ # 18 property distribution plots
│ ├── molecular_weight.png
│ ├── logp.png
│ ├── longest_aliph_carbon.png
│ └── ... (18 property plots)
└── conformers/ # OpenEye OMEGA conformer output (.oeb, logs)
Key result files:
CHEMBL2971_ID*.csv— the final curated dataset (one row per molecule, with tokens, properties,distribution_success/distribution_failures,scaffold_smiles, andsplitcolumns).vocab.json/vocab_curated.json— the base and curated vocabularies (see Vocabulary handling).split_report.json/split_report.png— the split report card and figure.distributions/— one PNG per molecular property.conformers/— the OpenEye OMEGA conformer output.
View trimmed execution log
2026-07-08 15:31:14 | [ 3B9CE574 ] | INFO | Starting pipe.
2026-07-08 15:31:14 | [ 3B9CE574 ] | INFO | Input (CHEMBL2971): ChEMBL ID from ChEMBL ID: CHEMBL2971 (0 rows)
2026-07-08 15:31:14 | [ 3B9CE574 ] | INFO | [1/8] Starting source.
2026-07-08 15:31:14 | [ SQL ] | INFO |
==================================================
| Fetching activity data for CHEMBL2971
| Database: ./data/chembl/36.db
==================================================
2026-07-08 15:31:14 | [ SQL ] | INFO | Total entries available: 56207
2026-07-08 15:31:15 | [ SQL ] | INFO | 56207 entries retrieved.
2026-07-08 15:31:15 | [ 3B9CE574 ] | INFO | [1/8] Finished source | success | 56207 rows | 0.86s
2026-07-08 15:31:15 | [ 3B9CE574 ] | INFO | [2/8] Starting chembl.
2026-07-08 15:31:15 | [ CHEMBL ] | INFO | RAW | Configured curation is optimal for CHEMBL2971.
2026-07-08 15:31:15 | [ CHEMBL ] | INFO | RAW
| standard_type | assay_type | target_organism | bao_format | assay_format | entries | % |
|:--------------|:-----------|:----------------|:------------|:-------------|:------------|---:|
| IC50 | B | Homo sapiens | BAO_0000357 | protein | 14859/56207 | 26 |
| k_off | B | Homo sapiens | BAO_0000357 | protein | 11913/56207 | 21 |
| kon | B | Homo sapiens | BAO_0000357 | protein | 11913/56207 | 21 |
| Ki | B | Homo sapiens | BAO_0000357 | protein | 3704/56207 | 7 |
| IC50 | B | Homo sapiens | BAO_0000219 | cell | 2605/56207 | 5 |
2026-07-08 15:31:15 | [ CHEMBL ] | INFO | VALID
| standard_type | assay_type | target_organism | bao_format | assay_format | entries | % |
|:--------------|:-----------|:----------------|:------------|:-------------|:-----------|---:|
| IC50 | B | Homo sapiens | BAO_0000357 | protein | 9646/14577 | 66 |
| Ki | B | Homo sapiens | BAO_0000357 | protein | 2120/14577 | 15 |
| IC50 | B | Homo sapiens | BAO_0000219 | cell | 1217/14577 | 8 |
2026-07-08 15:31:16 | [ CHEMBL ] | INFO | Conditional curation: 9642/56207.
2026-07-08 15:31:16 | [ CHEMBL ] | INFO | Standardization: IC50 → pIC50 (log10 conversion).
2026-07-08 15:31:16 | [ CHEMBL ] | WARNING | 25 suspicious SMILES detected. mistakes_only=True, error_margin=0.0001
2026-07-08 15:31:24 | [ CHEMBL ] | INFO | Removed 398 (canonical) SMILES entries with std > 0.5 or range > 0.5.
2026-07-08 15:31:24 | [ CHEMBL ] | INFO | Aggregated 530 duplicate SMILES entries. Final: 8060/9592 rows.
2026-07-08 15:31:25 | [ 3B9CE574 ] | INFO | [2/8] Finished chembl | success | 8060 rows | 9.53s
2026-07-08 15:31:25 | [ 3B9CE574 ] | INFO | [3/8] Starting curate.
2026-07-08 15:31:25 | [ CURATE ] | INFO | Processing 8,060 molecules with 19 processes (19 chunks of 425).
2026-07-08 15:31:25 | [ CURATE ] | INFO | Molecular curation: 8052/8060 successful. drop_na=True
2026-07-08 15:31:28 | [ CHEMBL ] | INFO | Aggregated 42 duplicate SMILES entries. Final: 7970/8052 rows.
2026-07-08 15:31:28 | [ 3B9CE574 ] | INFO | [3/8] Finished curate | success | 7970 rows | 3.77s
2026-07-08 15:31:28 | [ 3B9CE574 ] | INFO | [4/8] Starting tokens.
2026-07-08 15:31:28 | [ TOKENS ] | INFO | Vocabulary constructed: 48 tokens
2026-07-08 15:31:28 | [ TOKENS ] | INFO | Tokenized 7970 sequences.
2026-07-08 15:31:28 | [ 3B9CE574 ] | INFO | [4/8] Finished tokens | success | 7970 rows | 0.04s
2026-07-08 15:31:28 | [ 3B9CE574 ] | INFO | [5/8] Starting distributions.
2026-07-08 15:31:29 | [DISTRIBUTIONS] | INFO | INITIAL DISTRIBUTION
| property | count | mean | std | min | max | symmetric | normal |
|:---------------------|------:|-------:|------:|-------:|-------:|:----------|:-------|
| num_atoms | 7970 | 31.69 | 5.61 | 11 | 64 | ✓ | ✗ |
| num_rings | 7970 | 4.42 | 0.90 | 2 | 8 | ✓ | ✗ |
| size_largest_ring | 7970 | 6.21 | 1.49 | 6 | 19 | ✗ | ✗ |
| num_tokens | 7970 | 56.16 | 10.84 | 19 | 105 | ✓ | ✗ |
| longest_aliph_carbon | 7970 | 0.83 | 0.90 | 0 | 10 | ✗ | ✗ |
| molecular_weight | 7970 | 440.25 | 80.07 | 160.18 | 927.92 | ✓ | ✗ |
| logp | 7970 | 3.22 | 1.13 | -0.40 | 9.88 | ✓ | ✗ |
| tpsa | 7970 | 108.57 | 26.20 | 13.59 | 271.58 | ✓ | ✗ |
| ... (18 properties) | | | | | | | |
2026-07-08 15:32:03 | [DISTRIBUTIONS] | INFO | FILTER RESULTS
| property | passed | removed | pass_rate | thresholds |
|:---------------------|:----------|--------:|:----------|:----------------------------------|
| num_h_donors | 7340/7970 | 630 | 92.1% | ≥μ-2.0σ (0.23) , ≤μ+2.0σ (3.95) |
| num_atoms | 7535/7970 | 435 | 94.5% | ≥μ-2.0σ (20.47), ≤μ+2.0σ (42.91) |
| molecular_weight | 7558/7970 | 412 | 94.8% | ≥μ-2.0σ (280.11), ≤μ+2.0σ (600.39)|
| logp | 7627/7970 | 343 | 95.7% | ≥μ-2.0σ (0.96) , ≤μ+2.0σ (5.48) |
| longest_aliph_carbon | 7747/7970 | 223 | 97.2% | ≥μ-2.0σ (-0.97), ≤μ+2.0σ (2.63) |
| size_largest_ring | 7844/7970 | 126 | 98.4% | ≥μ-2.0σ (3.23) , ≤μ+2.0σ (9.19) |
2026-07-08 15:32:03 | [DISTRIBUTIONS] | INFO | Distribution curation: 7970 → 5292 retained (removed 2678, 33.6%) | property 2678, token 0
2026-07-08 15:32:03 | [ 3B9CE574 ] | INFO | [5/8] Finished distributions | success | 5292 rows | 34.72s
2026-07-08 15:32:03 | [ 3B9CE574 ] | INFO | [6/8] Starting confs.
2026-07-08 15:32:03 | [ OEO ] | INFO | Starting OMEGA generation for 5292 SMILES
2026-07-08 15:32:31 | [ OEO ] | INFO | Progress: 2074/5292 (39.2%) | Rate: 73.61 mol/s | ETA: 43.7s
2026-07-08 15:33:05 | [ OEO ] | INFO | Progress: 5292/5292 (100.0%) | Rate: 85.49 mol/s | Elapsed: 1.0m
2026-07-08 15:33:06 | [ OEO ] | INFO | OMEGA generation complete: 5259/5292 succeeded
2026-07-08 15:33:06 | [ CONFS-OEO ] | INFO | Conformer generation complete: 5259 succeeded, 33 failed, avg 92.8 conformers/molecule
2026-07-08 15:33:06 | [ CONFS-OEO ] | WARNING | Failure summary:
| Status | count |
|:---------------------------------------------------|------:|
| Failed due to unspecified stereochemistry | 6 |
| Failed to build structure from CT | 22 |
| Force field setup failed due to missing parameters | 5 |
2026-07-08 15:33:06 | [ 3B9CE574 ] | INFO | [6/8] Finished confs | success | 5259 rows | 1m 3.3s
2026-07-08 15:33:06 | [ 3B9CE574 ] | INFO | [7/8] Starting scaffold.
2026-07-08 15:33:07 | [ SCAFFOLD ] | INFO | Scaffold computation: 5259/5259 succeeded (0 acyclic, 0 failed).
2026-07-08 15:33:07 | [ 3B9CE574 ] | INFO | [7/8] Finished scaffold | success | 5259 rows | 0.51s
2026-07-08 15:33:07 | [ 3B9CE574 ] | INFO | [8/8] Starting split.
2026-07-08 15:33:07 | [ SPLIT ] | INFO | Splitter (unit=scaffold, method=isolation): 5,259 molecules → 10% test / 10% val / 80% train.
2026-07-08 15:33:07 | [ SPLIT ] | INFO | 2,018 scaffold units (0 rows → train unconditionally).
2026-07-08 15:33:07 | [ SPLIT ] | INFO | Isolation scores computed for 2,018 units in 0.09s.
2026-07-08 15:33:07 | [ SPLIT ] | INFO | Split: train=4,206 val=527 test=526
2026-07-08 15:33:08 | [ SPLIT ] | INFO |
── Split Summary ──
| | train | val | test |
|:------------|---------:|--------:|--------:|
| n_molecules | 4206.000 | 527.000 | 526.000 |
| fraction | 0.800 | 0.100 | 0.100 |
| n_scaffolds | 1474.000 | 286.000 | 258.000 |
| entropy | 0.885 | 0.922 | 0.888 |
── NN Tanimoto Distance (mol-level, median) ──
| query \ ref | train | val | test |
|:--------------|--------:|------:|-------:|
| train | 0.17 | 0.616 | 0.719 |
| val | 0.298 | 0.198 | 0.465 |
| test | 0.616 | 0.431 | 0.212 |
2026-07-08 15:33:10 | [ SPLIT ] | INFO | [8/8] Finished split | success | 5259 rows | 2.70s
2026-07-08 15:33:10 | [ 3B9CE574 ] | INFO | [8/8] Pipeline completed | Total time: 1m 55.4s
- Pipeline Configuration
- Command-Line Interface
- Python API
- Plugin Development — the
scaffoldandsplitplugins used here