Truace
Subscribe

Benchmarking ensemble and deep learning models for drought-responsive gene prioritisation in rice using multi-omics data

Identifying drought-responsive genes is critical for climate-resilient crop breeding, yet computational gene-trait association often relies on single-omics or generalized genomic datasets. Optimal algorithm selection for high-dimensional, small multi-omics datasets also remains a persistent challenge. We present a machine learning framework that extends the QTG-Finder feature set by integrating transcriptomic and proteomic features to classify drought-responsive genes in rice. Using MBKBase annotations, we evalu…

By the Truvace desk2 min read1 sourcePeer-reviewedTRV-2026-1675· cite & verify
Benchmarking ensemble and deep learning models for drought-responsive gene prioritisation in rice using multi-omics data
"Dry Fields, Drought in Bogor Regency" by DMahendra is licensed under CC BY 2.0. To view a copy of this license, visit https://creativecommons.org/licenses/by/2.0/.

In brief

Researchers extended the QTG-Finder feature set with transcriptomic and proteomic features and trained classifiers to pick out 336 drought-responsive genes from 2,416 abiotic stress-associated rice genes annotated in MBKBase. They compared Random Forest, XGBoost, a one-dimensional convolutional neural network and a multilayer perceptron.

The work suggests that combining several types of molecular data can help narrow candidate genes for climate-resilient breeding, and that simpler tree-based models may suit small tabular datasets better than deep learning. The results come from a limited, imbalanced dataset and a very small holdout, so how well the approach would work on new genes or in breeding programmes remains unresolved.

Main points

  1. The dataset contained 336 drought-responsive genes among 2,416 abiotic stress-associated genes, leaving 2,080 as an operational background set with severe class imbalance.
  2. Tree-based ensemble methods outperformed the deep learning models, which showed early overfitting.
  3. The fine-tuned Random Forest reached recall 0.67 and AUC-ROC 0.87 under nested cross-validation.
  4. Ablation and feature importance analyses pointed to transcriptomics-derived features, especially expression distribution, magnitude and direction, as the largest contributors.

The gain

Adding transcriptomic and proteomic features to a genomics-based classifier raised recall for identifying drought-responsive rice genes from 0.47 to 0.67 in a small multi-omics dataset.

The rundown

The framework builds on QTG-Finder by adding transcriptomic and proteomic layers to the genomic features, aiming to overcome reliance on single-omics or generalized genomic data in computational gene-trait association. The authors also describe algorithm selection for high-dimensional, small multi-omics datasets as a persistent challenge.

Four algorithms were tested while the authors dealt with class imbalance and checked model stability. Random Forest and XGBoost outperformed the one-dimensional CNN and multilayer perceptron, the latter pair showing early overfitting in this dataset.

On a small internal holdout of 10 genes, repeated evaluations gave mean accuracy of 0.90 and mean recall of 0.63, which the authors report only as a consistency check. Functional enrichment analysis was used to support the biological relevance of the prioritised genes.

Sources

  1. Peer-reviewedPLOS One2026-10-09

The debate