Identification and functional annotation of potential biomarkers associated with thalassemia using machine learning-based knowledge discovery

Abstract

Thalassemia, a hereditary blood disorder, causes abnormal hemoglobin production—alpha- and beta-thalassemia are its variants. This leads to decreased hemoglobin levels and accounted for 16,800 deaths in 2015, affecting 1.5% of the global population. Diagnosis involves blood tests and genetic screening, but many severe cases go undiagnosed due to limited registries and screening, resulting in high mortality. Our work suggests using gene expression profiling and machine learning to identify biomarkers for thalassemia. Using an Isolation Forest algorithm, we found 72 anomalous genes. Validation showed significant terms like cytoplasmic translation and apoptosis, indicating potential pathways for thalassemia. We also found genes related to iron homeostasis, linking oxidative stress and apoptosis to thalassemia. Comparing with another study, we found common processes. Five genes identified in AmiGO are up-regulated in thalassemia and could be biomarkers due to their abnormal expression and biological role. This highlights the potential of machine learning in refining diagnosis and understanding molecular pathways for better patient management, calling for further research.