A comprehensive targeted panel of 295 genes: Unveiling key disease initiating and transformative biomarkers in multiple myeloma

Vivek Ruhela; Ritu Gupta; Rupin Oberoi; Anubha Gupta

doi:10.1016/j.compbiomed.2025.110619

A comprehensive targeted panel of 295 genes: Unveiling key disease initiating and transformative biomarkers in multiple myeloma

Comput Biol Med. 2025 Jul 4;196(Pt A):110619. doi: 10.1016/j.compbiomed.2025.110619. Online ahead of print.

Authors

Vivek Ruhela¹, Ritu Gupta², Rupin Oberoi³, Anubha Gupta⁴

Affiliations

¹ SBILab, Deptt. of ECE & Centre of Excellence in Healthcare, Indraprastha Institute of Information Technology Delhi (IIIT-Delhi), India; Department of Computational Biology, Indraprastha Institute of Information Technology Delhi (IIIT-Delhi), India.
² Laboratory Oncology Unit, Dr.B.R.A.IRCH, All India Institute of Medical Sciences, Delhi, India.
³ SBILab, Deptt. of ECE & Centre of Excellence in Healthcare, Indraprastha Institute of Information Technology Delhi (IIIT-Delhi), India.
⁴ SBILab, Deptt. of ECE & Centre of Excellence in Healthcare, Indraprastha Institute of Information Technology Delhi (IIIT-Delhi), India. Electronic address: anubha@iiitd.ac.in.

PMID: 40617081
DOI: 10.1016/j.compbiomed.2025.110619

Abstract

Background: Multiple myeloma (MM) is a hematological malignancy that progresses from a benign precursor stage known as Monoclonal Gammopathy of Undetermined Significance (MGUS). Distinguishing MM from MGUS at the molecular level by identifying key biomarkers, genomic alterations, and gene interactions is critical for early detection and deeper insight into MM pathogenesis.

Methods: We have developed an advanced genomics domain-rooted AI-workflow that combines the traditional statistical mutation profiling methods with the proposed BIO-DGI (Bio-Inspired Graph Network Learning-based Gene-Gene Interaction) attention-based deep learning architecture exploiting gene-gene interaction. Our proposed framework utilizes multiple variant profiles including SNVs and CNVs extracted from WES data and SVs extracted from WGS data. Rigorous post-hoc validation including ShAP analysis, community analysis, survival analysis, Geo2R validation, and pathway enrichment analysis are utilized to eventually design the panel.

Results: BIO-DGI outperformed traditional machine learning and deep learning methods on quantitative metrics and identified the highest number of MM-relevant genes in post-hoc analysis. ShAP analysis of gene SNV profiles, community analysis of the disease-specific gene-gene graph, survival analysis of SNVs, CNVs, and SVs led to the design of 295 gene-panel for multiple myeloma. The pathway enrichment analysis confirmed strong association of our gene-panel with the MM-related biological pathways.

Conclusion: This study presents a comprehensive framework combining bio-inspired graph learning, multi-variant genomic profiling, post-hoc interpretability, and survival-driven clinical validation to advance biomarker discovery in multiple myeloma. By integrating exomic variants with network-based gene-gene interactions through the novel BIO-DGI model, we developed a clinically curated 295-gene panel for MM.

Keywords: AI in cancer; GCN in cancer; Gene-panel; Hematological malignancy; MGUS; Multiple myeloma; ShAP analysis.