Truace
Subscribe

DeepFusion-ACSM: an interpretable hybrid ML-DL ensemble model for anticancer small-molecule activity prediction

Cancer remains one of the leading causes of mortality worldwide, and the discovery of novel anticancer drugs continues to be a costly and time-consuming process. Although machine learning (ML) and deep learning (DL) have significantly advanced computational drug discovery, many existing predictive models rely on a single learning algorithm and often fail to generalize across diverse molecular descriptors, limiting their predictive reliability. In this study, we propose an interpretable hybrid ML-DL ensemble fram…

By the Truvace desk1 min read1 sourcePeer-reviewedTRV-2026-1723· cite & verify
DeepFusion-ACSM: an interpretable hybrid ML-DL ensemble model for anticancer small-molecule activity prediction
Molecular modeling by Dr. Jason Kang (Photographer). Public domain

In brief

Researchers built a computational framework that fuses many machine learning and deep learning models to predict whether small molecules have anticancer activity. They trained 102 ML models and 12 DL models, then combined their out-of-fold predictions through eight ensemble strategies.

The result matters because drug discovery is described as costly and slow, and single-algorithm predictors often generalize poorly. The evidence is computational, with a small external check on 10 approved compounds, so real-world drug discovery benefit has not been shown in the supplied text.

Main points

  1. The best ensemble combined XGBoost, Improved Deep MLP and TabNet, trained on the 2D descriptor dataset with mutual information feature selection.
  2. It reached macro-recall 0.83, accuracy 0.83 and ROC-AUC 0.90, outperforming individual models.
  3. External validation correctly labeled 8 of 10 FDA-approved compounds with known anticancer or non-anticancer status.
  4. The framework was applied to 1200 FDA-approved drugs from the ZINC database for prospective virtual screening.

The gain

Combining XGBoost, a deep MLP and TabNet in an ensemble identified anticancer small molecules more accurately than individual ML or DL models, which could speed early-stage drug discovery.

The rundown

The study used 17 algorithms across two molecular descriptor datasets, a 2D set and a Hybrid set, with three feature selection strategies. Twelve deep learning models based on Improved Deep MLP, FT-Transformer and TabNet were optimized with five-fold cross-validation.

Out-of-fold predictions from the optimized models were merged by eight ensemble strategies to form ML, DL and hybrid ML-DL ensembles. The hybrid ensemble performed best.

For prospective screening, 1200 FDA-approved drugs from the ZINC database were ranked by ensemble prediction probability to prioritize candidates. The authors describe the framework as reproducible, interpretable and scalable.

Sources

  1. Peer-reviewedJournal of Computer-Aided Molecular Design2026-10-08

The debate