AML-1264: Proteome-Based Clustering Reveals Distinct Molecular Subtypes in Acute Myeloid Leukemia: A BEAT AML Study

Acute myeloid leukemia (AML) is a biologically heterogeneous malignancy characterized by diverse molecular and clinical profiles. While genomic subtyping has improved prognostication, proteomic stratification remains underexplored. Leveraging high-dimensional proteomic data from the BEAT AML initiative, we applied unsupervised machine learning to identify clinically meaningful subtypes based on protein expression. We analyzed peptide spectral count data from 218 samples from patients with AML in the CPTAC3-PNNL BEAT AML proteomics dataset. After aggregating counts to the gene level, we selected the top 1,000 most highly expressed proteins. Principal component analysis (PCA) was performed for dimensionality reduction, and K-means clustering ( K = 3) was applied to identify molecular subtypes. Pairwise t tests were conducted to identify top differentially expressed proteins among clusters. Patient sample identifiers were linked to metadata to support future integration with survival and treatment response data. PCA revealed clear separation between three distinct proteomic clusters. Statistical analysis identified 25 proteins with significant differential expression across clusters ( P < 0.001). Heatmap visualization confirmed distinct expression profiles between subtypes. These clusters suggest unique underlying biology, with preliminary annotation indicating differences in immune regulation and metabolic pathway activity. Ongoing work aims to integrate these clusters with clinical outcomes to assess prognostic significance. Unsupervised clustering of AML proteomic data revealed three reproducible molecular subtypes with distinct protein expression signatures. This proteome-based stratification offers a complementary lens to genomic profiling and may help refine AML risk classification and therapeutic targeting. Future integration with clinical outcome data will assess the utility of these subtypes for prognostication and personalized treatment planning.