Nonnegative Matrix Factorization of Count Data with the Shifted Log Transform

Learn advanced techniques for applying nonnegative matrix factorization (NMF) to count data using the shifted log transform in this 43-minute conference talk from the Computational Genomics Summer Institute. Explore how NMF algorithms can be enhanced to better handle count data structures commonly found in genomics applications, with particular focus on the mathematical foundations and practical implementations of the shifted log transform approach. Discover how these improved NMF methods significantly enhance topic model fits and can be applied to dissect transcriptional heterogeneity in single-cell RNA-seq data. Examine the theoretical underpinnings of non-negative matrix factorization for learning parts-based representations of data, and understand how generalized binary covariance decomposition extends these concepts for analyzing tumor transcriptional heterogeneity. Gain insights into cutting-edge computational methods that bridge classical matrix factorization techniques with modern genomics data analysis challenges, supported by recent research developments in the field.