Non-Negative Matrix Factorization Of The Human Cellcensus
Location
Hager-Lubbers Exhibition Hall
Description
PURPOSE: Non-negative Matrix Factorization (NMF) is a powerful dimensionality reduction technique widely used in transcriptomic analysis. In this study, we apply NMF to human Cell Census transcriptomes at a rank of 200, leveraging publicly available implementations in RcppML. NMF decomposes the gene expression matrix A into three components: W (transcripts by factors), D (a diagonal scaling vector), and H (factors by cells). By analyzing the W and H matrices, we investigate patterns of gene expression and cell-type associations across 15 million cells. METHODS AND MATERIALS: Through metadata-driven annotation, we identify factor enrichments in specific cell types, enabling the characterization of gene-cell type relationships. RESULTS: This approach facilitates the discovery of biologically relevant factors and potential regulatory mechanisms. CONCLUSIONS: Our findings have implications for single-cell transcriptomics, enhancing our understanding of cellular heterogeneity and providing insights into disease mechanisms and therapeutic targets. Future applications of this work include refining cell-type classification, improving biomarker discovery, and integrating NMF-derived features into predictive modeling for precision medicine.
Non-Negative Matrix Factorization Of The Human Cellcensus
Hager-Lubbers Exhibition Hall
PURPOSE: Non-negative Matrix Factorization (NMF) is a powerful dimensionality reduction technique widely used in transcriptomic analysis. In this study, we apply NMF to human Cell Census transcriptomes at a rank of 200, leveraging publicly available implementations in RcppML. NMF decomposes the gene expression matrix A into three components: W (transcripts by factors), D (a diagonal scaling vector), and H (factors by cells). By analyzing the W and H matrices, we investigate patterns of gene expression and cell-type associations across 15 million cells. METHODS AND MATERIALS: Through metadata-driven annotation, we identify factor enrichments in specific cell types, enabling the characterization of gene-cell type relationships. RESULTS: This approach facilitates the discovery of biologically relevant factors and potential regulatory mechanisms. CONCLUSIONS: Our findings have implications for single-cell transcriptomics, enhancing our understanding of cellular heterogeneity and providing insights into disease mechanisms and therapeutic targets. Future applications of this work include refining cell-type classification, improving biomarker discovery, and integrating NMF-derived features into predictive modeling for precision medicine.