Dropout-Aware Weighted NMF on scRNA-seq Data

This is the final project for CS2952Q Robust Algorithms for Machine Learning.

Non-negative matrix factorization (NMF) aims to infer low-dimensional structure from high-dimensional genomic data to understand and visualize complex biological processes. In recent years, the rapid development of single-cell RNA sequencing (scRNA-seq) enables a single-cell level of gene profiling. Previous studies have applied NMF on scRNA-seq data to obtain a finer resolution of biological system understanding, but the high sparsity and the complicated structure of dropouts (i.e., zero values) in scRNA-seq data pose challenges to NMF optimization. To address this problem, in this study, we propose a dropout-aware weighted non-negative matrix factorization (DA-WNMF) to deal with the high sparsity based on Bayesian factorization. On experimental scRNA-seq data, DA-WNMF outperforms previous works when the data is not too sparse (e.g., has around 80% zeros) but fails for very-sparse data (i.e., has more than 90% zeros).