Machine Learning Reveals Two Distinct Transcriptomics Subtypes of Acute Myeloid Leukaemia with Differences in Disease Outcomes and Genetic Landscape

Author(s): Panji Nkhoma, Doris Kafita, Kevin Dzobo, Sinkala Musalula.

Background: Accurately subtyping diseases, particularly in cancer research, is crucial for enhancing the precision of treatment decisions and improving outcomes across various cancers, including hematological malignancies like acute myeloid leukaemia (AML). Methods: Consequently, we utilised an unsupervised K-means clustering on transcriptomics data from 173 acute myeloid leukaemia samples profiled by The Cancer Genome Atlas (TCGA). In our analysis, we categorised patients into two distinct groups: Subtype-1, comprising 68 individuals, and subtype-2, encompassing 105 individuals. Results: Analysis revealed that individuals within subtype-2 experienced a markedly prolonged period of disease-free survival compared to those in subtype-1, as evidenced by the Log-rank test (p = 0.00273). Furthermore, it was observed that patients in subtype-1 presented with elevated white blood cell counts, suggesting a potential biomarker of disease progression within this subgroup. We also identified differentially expressed genes linked to poor survival, prognosis, and chemoresistance, involving pathways like Aminoacyl-tRNA biosynthesis, apoptosis, NF-kappa B and HIF-1, through bioinformatics analysis of subtype-1. Our findings show that AML patients categorised within subtype-1 exhibit a more aggressive form of the disease compared to those allocated to subtype-2. Conclusion: Consequently, these observations underscore the feasibility of employing subtype-specific precision treatments for AML patients, offering a tailored therapeutic approach based on the distinct disease characteristics of different patient subtypes.