A multi-target QSAR framework spanning dozens of breast cancer targets pairs ensemble models with a small in vitro check of its top picks
- BreastCAR trains several classical machine-learning models on public ChEMBL data for roughly fifty cancer-relevant targets, using fingerprints plus physicochemical descriptors.
- Each model predicts potency as a continuous value and as a three-tier activity class.
- Permutation controls and an applicability-domain check suggest the models learn real signal, and most test compounds sit inside the training domain.
- A panel of 13 benzylidene compounds was tested against MCF-7 cells to probe the top-ranked predictions.
💡 Points Based on the abstract
Permutation controls show only that the models beat noise, not that the split blocks analog leakage, which matters because ChEMBL series are dense with near neighbours. Unless splits were scaffold- or time-based, expect optimistic numbers. Thirteen benzylidene compounds in MCF-7 cells is a single-chemotype, cell-level check that cannot validate target-specific predictions. Per-target results against a plain Morgan-fingerprint random forest baseline would help.
BreastCAR: A Machine Learning Framework for Breast Cancer Drug Activity Prediction ↗