Multiple machine-learning tools identifying prognostic biomarkers for acute Myeloid Leukemia.

Area of Science:

• Oncology
• Bioinformatics
• Genetics

Background:

• Acute Myeloid Leukemia (AML) presents a significant challenge with its generally low survival rates post-treatment.
• There is a critical need for novel biomarkers to enhance the prognostic accuracy for AML patients.
• Machine learning approaches are increasingly vital in identifying potential biomarkers.

Purpose of the Study:

• To identify key genes associated with Acute Myeloid Leukemia (AML) prognosis using multiple machine learning models.
• To validate the diagnostic and prognostic value of identified genes in AML and across various cancers.
• To explore the correlation of these genes with AML immune subtypes and immune checkpoints.

Main Methods:

• Application of diverse machine learning algorithms including LASSO, SVM-RFE, RF, and XGBoost for gene screening.
• Utilizing the Cancer Genome Atlas (TCGA) for pan-cancer analysis to assess gene correlations.
• Performing verification analyses on different datasets to confirm the diagnostic utility of selected genes.

Main Results:

• Survival analysis identified 26 differentially expressed genes, with DNM1, MEIS1, and SUSD3 selected as key genes.
• Pan-cancer analysis revealed MEIS1 and DNM1 are highly expressed in AML, with MEIS1/SUSD3 as risk factors and DNM1 as a protective factor.
• The three key genes demonstrated significant associations with AML immune subtypes and immune checkpoints, and possess potential diagnostic value.

Conclusions:

• DNM1, MEIS1, and SUSD3 have been identified as promising prognostic biomarkers for Acute Myeloid Leukemia (AML) through the application of multiple machine learning techniques.
• These genes hold potential for improving patient outcomes and guiding therapeutic strategies in AML management.