Ensembling Deep Learning and Bioinformatics for Precision Treatment Recommendations: Application in Acute Myeloid Leukemia

The most common kind of acute leukemia, acute myeloid leukemia (AML), has poor clinical outcomes and inadequate therapy. Although there are many therapies for AML, AML patients still have low 5-year overall survival rates. This study aimed to identify key genes and pathways of AML, to predict the prognosis of AML patients, A total of 133 upregulated DEGs and 15 downregulated DEGs were identified. They were enriched in protein binding, insulin secretion regulation, and cancer transcriptional misregulation. The module analysis of the PPI network of DEGs identified ten genes (PCLO, SYT1, STXBP1, GPHN, SYNCRIP, PFN2, ENAH, PAIP1, HNRNPD, STXBP5). SYT1 and PCLO were overexpressed in AML patients and have great potential as prognostic factors and therapeutic targets for AML. Further, by Deep learning, we combined other factors, including age, gender, past malignancy, and prior therapy, to inference whether patients will survive within 6 years by deep learning, the (C td )-index was greater than 80%, which indicates DeepHit model can reliably predict the prognosis of AML patients.