Multi-Target Breast Cancer Activity Predictor

Enter a SMILES string to predict pIC50 and activity class across 53 validated breast cancer targets using ensemble ML (RF · XGB · LGB).

Try: Tamoxifen Lapatinib analogue Aspirin Caffeine
Running predictions across 53 targets…
–
Active Targets
pIC50 ≥ 7.0  ·  IC50 ≤ 100 nM
–
Moderate Targets
pIC50 5.0–7.0  ·  100 nM–10 µM
–
Inactive Targets
pIC50 < 5.0  ·  IC50 > 10 µM
Query Molecule
Physicochemical Properties
Activity Distribution
Top 15 Targets by Predicted pIC50
Download full prediction results
Detailed Predictions by Target
Target ↕ pIC50 ↕ IC50 ↕ Activity Confidence

Batch Prediction

Predict activity for multiple compounds simultaneously — one SMILES per line.

Processing batch…

Target Library

53 validated breast cancer targets · Training set statistics · Best-performing models per target

Target Total Active Moderate Inactive % Active Distribution Mean pIC50 Best Reg. Best Clf.
AKT1 3571 1671 1639 261 46.8%
6.9 XGB LGB
AKT2 1061 500 512 49 47.1%
6.92 LGB XGB
AKT3 304 159 128 17 52.3%
7.03 XGB SVM
ANDR 2292 627 1530 135 27.4%
6.43 XGB LGB
AROMATASE 1980 490 1200 290 24.7%
6.22 XGB XGB
ASK1 1535 1113 374 48 72.5%
7.43 XGB LGB
AURA 3027 1338 1501 188 44.2%
6.88 LGB XGB
BCL2 2431 1898 454 79 78.1%
7.82 RF LGB
BRAF 5555 3729 1614 212 67.1%
7.47 LGB XGB
BRD4 7813 2298 4085 1430 29.4%
6.19 XGB XGB
CA9 249 65 151 33 26.1%
6.23 LGB RF
CCND1 369 207 147 15 56.1%
7.09 LGB XGB
CCNE1 58 35 23 0 60.3%
7.23 LGB LGB
CDK4 3820 2101 1327 392 55.0%
6.98 XGB RF
CDK6 1898 985 845 68 51.9%
7.0 XGB LGB
CHK2 431 179 186 66 41.5%
6.42 RF LGB
CXCR4 910 447 433 30 49.1%
6.86 XGB XGB
EGFR 12377 6493 4893 991 52.5%
7.03 XGB XGB
ERBB4 243 147 89 7 60.5%
7.22 XGB XGB
FGFR 3220 1897 1154 169 58.9%
7.2 LGB RF
HDAC1 9811 3606 5385 820 36.8%
6.59 LGB XGB
HDAC6 7307 4163 2824 320 57.0%
6.99 XGB XGB
HER2 4041 2185 1566 290 54.1%
6.9 XGB LGB
HER3 57 31 16 10 54.4%
6.86 XGB XGB
JAK2 11273 6002 4488 783 53.2%
7.02 XGB LGB
KRAS 10417 5155 4661 601 49.5%
6.97 LGB XGB
LDHA 717 89 383 245 12.4%
5.76 RF LGB
MAPK14 4842 2403 2222 217 49.6%
6.95 LGB LGB
MDM2 4444 3201 1011 232 72.0%
7.61 LGB XGB
MMP14 667 155 383 129 23.2%
6.03 LGB SVM
MMP2 3257 1326 1274 657 40.7%
6.51 XGB LGB
MMP9 2081 1081 692 308 51.9%
6.88 XGB XGB
MTOR 4147 2533 1500 114 61.1%
7.25 XGB XGB
NFKB1 137 18 104 15 13.1%
5.9 RF RF
NOTC1 112 82 26 4 73.2%
7.69 RF RF
OEST_A 3911 1970 1576 365 50.4%
7.01 XGB XGB
OEST_B 1780 731 656 393 41.1%
6.43 XGB RF
PARP_1 4107 2534 1359 214 61.7%
7.19 LGB XGB
PARP_10 197 3 160 34 1.5%
5.51 SVM SVM
PD1L1 3842 2390 1206 246 62.2%
7.51 LGB XGB
PK3CA 7635 2744 4224 667 35.9%
6.57 XGB XGB
PPAR_GAMMA 1558 587 832 139 37.7%
6.54 XGB XGB
PROGESTERONE 1452 580 854 18 39.9%
6.8 XGB XGB
RASN 1147 289 846 12 25.2%
6.26 RF RF
S1PR1 987 586 360 41 59.4%
7.27 XGB RF
STAT3 1050 68 622 360 6.5%
5.43 XGB RF
TGFR1 2692 1658 972 62 61.6%
7.24 XGB XGB
TGFR2 427 46 349 32 10.8%
5.92 XGB XGB
TLR4 122 7 74 41 5.7%
5.45 LGB MLP
TNF_ALPHA 1349 128 866 355 9.5%
5.66 XGB RF
TP53 283 208 67 8 73.5%
8.35 LGB MLP
VGFR2 8627 3850 4262 515 44.6%
6.81 XGB XGB
WEE1 1019 713 288 18 70.0%
7.48 XGB XGB

BreastCAR — a breast cancer drug discovery suite

A machine-learning platform that scores any small molecule against 53 validated breast-cancer targets in seconds, delivering both a continuous potency estimate (pIC50) and an Active / Moderate / Inactive class call.

53
Targets
106
Trained models
233,975
Bioactivities
5
Algorithms
2048
ECFP4 bits
1
SMILES
User input molecule
→
2
Featurise
Morgan FP + RDKit descriptors
→
3
Score
53 regressors + 53 classifiers
→
4
Rank
Top targets by predicted pIC50
→
5
Report
Table, chart, properties

Methodology

  • FeaturesMorgan (ECFP4) fingerprint — 2048 bits, radius 2 — concatenated with 13 physicochemical RDKit descriptors (MolLogP, MolMR, TPSA, HBA/HBD, RotBonds, RingCount, AromaticRings, FractionCSP3, HeavyAtomCount, HeteroAtoms, BalabanJ, BertzCT).
  • AlgorithmsRandom Forest, XGBoost, LightGBM, SVM, MLP — benchmarked per target; best performer retained.
  • ValidationStratified 80/20 split, 5-fold cross-validation, y-randomisation, and Bemis–Murcko scaffold split.
  • ApplicabilityTanimoto-similarity threshold vs the training set; out-of-domain predictions are flagged.
  • LabelsActive if pIC50 ≥ 7, Moderate if ≥ 5, else Inactive.

Data source

BindingDB — curated IC50 and Ki measurements against human targets implicated in breast-cancer biology: hormone receptors, cyclin-dependent kinases, DNA-damage response, growth-factor signalling, and the tumour micro-environment.

Records were standardised, salt-stripped, de-duplicated, and converted to pIC50. Final dataset: 233,975 bioactivity measurements across 53 targets.

API

POST /predict · POST /predict/batch · GET /targets · GET /health

How to use

  • SinglePaste one SMILES → predictions for all 53 targets ranked by pIC50, with 2D structure and drug-likeness.
  • BatchUpload a SMILES list, download the full result table.
  • ExploreBrowse training-set composition per target with the winning algorithm.

Disclaimer

BreastCAR provides computational predictions for research triage and hypothesis generation. Results are not a substitute for experimental validation. Predictions on compounds outside the applicability domain of the training data should be interpreted with caution.

Draw molecule

Sketch a structure — click Use SMILES to send it to the predictor.