pyNVR: investigating factors affecting feature selection from scRNA-seq data for lineage reconstruction

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) enables pseudotemporal trajectory reconstruction by analyzing transitioning cell states.
• scRNA-seq data presents challenges like sparsity, zero inflation, and technical variation, requiring high-quality feature selection for downstream analyses.
• Existing unsupervised feature selection algorithms have differential performance that remains largely unaddressed.

Purpose of the Study:

• To implement and evaluate the Neighborhood Variance Ratio (NVR) feature selection method as a Python package.
• To compare the performance of NVR against other unsupervised feature selection algorithms.
• To investigate the influence of quantifiable dataset properties on algorithm performance.

Main Methods:

• Implementation of the Neighborhood Variance Ratio (NVR) as a Python package.
• Comparison of NVR with existing unsupervised feature selection algorithms, including dpFeature.
• Analysis of dataset properties to understand their impact on feature selection algorithm performance.

Main Results:

• The pyNVR package demonstrates substantial performance improvements in feature selection.
• Striking differences were observed in the features selected by NVR compared to algorithms like dpFeature.
• Quantifiable dataset properties were shown to have predictable effects on the performance of feature selection algorithms.

Conclusions:

• NVR is an effective feature selection method for scRNA-seq data, offering improved performance.
• Understanding dataset properties is crucial for selecting appropriate feature selection algorithms.
• The pyNVR package provides a valuable tool for scRNA-seq data analysis.