Introducing feature identification and refinement engine (FIRE) for identifying consistent and informative gene signature.

Identifying disease-specific molecular signatures from omics data remains a major challenge in biomedical research. In the case of glioblastoma, one of the most aggressive forms of brain tumors, clinical management is complicated by molecular heterogeneity and the lack of a consistent, reproducible signature. Despite extensive research, many studies fall short due to high data variance and complex biological patterns. To address this critical gap, we present FIRE (Feature Identification and Refinement Engine), a novel machine learning (ML)-driven framework designed for integrative analysis and feature refinement. FIRE incorporates data merging and an ensemble ML approach capable of detecting both linear and non-linear patterns across diverse datasets. This enables robust extraction of biologically meaningful features from complex, heterogeneous omics data. Applying FIRE to glioblastoma datasets, we identified 33 genes that consistently distinguish glioblastoma from control samples. A literature survey further revealed several of these genes associated with established cancer hallmarks. The robustness of FIRE was established through repeated cross-validation across multiple independent datasets, demonstrating superior predictive performance compared to existing glioblastoma signatures, projecting it as a promising tool applicable to other diseases.