Related Experiment Video
Updated: May 16, 2026
Generating the Transcriptional Regulation View of Transcriptomic Features for Prediction Task and Dark Biomarker Detection on Small Datasets
Published on: March 1, 2024
Hybrid Feature Selection-Based Machine Learning and Deep Learning Framework for Biomarker Prediction From RNA-seq
Srilekha Anumulapuri1,2, Jhansi Venkata Nagamani Josyula1,2, Agiesh Kumar Balakrishna Pillai3
1Department of Applied Biology & Centre for Information Technology, CSIR-Indian Institute of Chemical Technology, Hyderabad, India.
Background:
Severe dengue (SD) represents a life-threatening progression of dengue virus infection. Early identification of patients at risk of transitioning from dengue fever (DF) to SD remains a major clinical challenge. Unraveling the transcriptomic changes underlying this progression may aid in developing timely therapeutic interventions.
Methods:
RNA-seq datasets comprising 103 samples (62 SD and 41 DF) were retrieved from the GEO repository. Following normalization using DESeq2, differentially expressed genes (DEGs) were identified between the 2 disease stages. Functional enrichment analysis was performed to uncover dysregulated biological processes. A hybrid computational framework combining classical machine learning (Logistic Regression, Support Vector Machine, and Random Forest) and deep learning models (Artificial Neural Network, Convolutional Neural Network, and Transformer-based architectures) were applied to classify SD and DF samples. Model performance was evaluated using ROC-AUC and balanced accuracy metrics.
Results:
Differential expression analysis identified 55 significantly dysregulated genes distinguishing severe dengue from dengue fever. These genes were enriched in pathways related to metal ion homeostasis, platelet signaling, ferroptosis, and oxidative stress. Among multiple machine learning and deep learning models, the Transformer-CNN achieved the best performance (test AUC = 0.85; balanced accuracy = 0.89). SHAP-based interpretation highlighted ILDR2, TCP1, HNRNPUL1, SEC14L5, ATP2C2, LOXL3, ACVRL1, STEAP3, and ST8SSIA5 as key discriminative features. Integrated network analyses further implicated coordinated regulation of iron metabolism, calcium signaling, and platelet dysfunction in severe dengue.
Conclusion:
This study integrates RNA-seq and hybrid Artificial Intelligence modeling to identify transcriptomic signatures associated with dengue severity. The study highlights candidate genes and pathways that provide a hypothesis-generating foundation; further increasing the sample size and experimental validation will support early risk stratification in severe dengue.