scPNMF

scPNMF selects informative genes from single-cell RNA sequencing (scRNA-seq) data by applying a projective non-negative matrix factorization (PNMF) framework to identify gene bases that capture cell-type–relevant variation.

Key Features:

• Unsupervised Gene Selection: Performs unsupervised identification of informative genes without requiring predefined labels or categories.
• Enhanced Cell Type Distinction: Selects gene bases that improve separation and distinction among different cell types.
• Alignment with Targeted Gene Profiling Data: Enables alignment of targeted gene profiling datasets with reference scRNA-seq data in a low-dimensional space to support cell-type prediction.
• Technical Innovations: Extends Non-negative Matrix Factorization (NMF) by employing projective NMF with altered initialization procedures and an additional basis selection step to choose informative bases.
• Performance Superiority: Empirically outperforms existing state-of-the-art gene selection methods across multiple scRNA-seq datasets.

Scientific Applications:

• Targeted Gene Profiling Experiment Design: Guides selection of genes for targeted gene profiling experiments to focus assays on informative transcripts.
• Cell-Type Annotation: Supports cell-type annotation of targeted profiling data by integrating with reference scRNA-seq datasets in a shared low-dimensional representation.

Methodology:

Modifies the conventional PNMF algorithm by incorporating altered initialization procedures and an additional basis selection step and projects data into a low-dimensional space for alignment between targeted profiling and reference scRNA-seq datasets.