Acute myeloid leukemia (AML) is a blood cancer with high heterogeneity and stratified as M0-M7 subtypes in the French-American-British (FAB) diagnosis system. Improved diagnosis with leverage of key molecular inputs will assist precisive medicine. Through deep-analyzing the transcriptomic data and mutations of AML, we report that a modern clustering algorithm, t-distributed Stochastic Neighbor Embedding (t-SNE), successfully demarcates M2, M3 and M5 territories while M4 bias to M5 and M0 & M1 bias to M2, consistent with the traditional FAB classification. Combining with mutation profiles, the results show that top recurrent AML mutations were unbiasedly allocated into M2 and M5 territories, indicating the t-SNE instructed transcriptomic stratification profoundly outperforms mutation profiling in the FAB system. Further functional data mining prioritizes several myeloid-specific genes as potential regulators of AML progression and treatment by Venetoclax, a BCL2 inhibitor. Among them two encode membrane proteins, LILRB4 and LRRC25, which could be utilized as cell surface biomarkers for monocytic AML or for innovative immuno-therapy candidates in future. In summary, our deep functional data-mining analysis warrants several unappreciated immune signaling-encoding genes as novel diagnostic biomarkers and potential therapeutic targets.
Keywords: Acute myeloid leukemia; Biomarker; Stratification and prognosis; Transcriptome; Venetoclax; t-SNE.
© 2022. The Author(s).