Accurate sample deconvolution of pooled snRNA-seq using sex-dependent gene expression patterns.

Area of Science:

• Genomics
• Computational Biology
• Neuroscience

Background:

• Single-nucleus RNA sequencing (snRNA-seq) provides high-resolution gene expression data but faces cost and technical limitations, often necessitating sample pooling.
• Pooling samples in snRNA-seq sacrifices individual sample-level data and increases experimental costs.
• Developing methods for sample deconvolution in pooled snRNA-seq is crucial for enhancing data throughput and analytical power.

Purpose of the Study:

• To demonstrate the feasibility of deconvoluting pooled snRNA-seq data using inherent biological features.
• To leverage sex-dependent gene expression patterns for identifying individual sample origins.
• To benchmark machine learning models for accurate cell sex classification and sample deconvolution.

Main Methods:

• Utilized previously published snRNA-seq data from the rat ventral tegmental area.
• Trained various machine learning models to classify cell sex based on differentially expressed genes between male and female rats.
• Compared classification accuracy using sex-dependent genes versus solely sex chromosome genes.

Main Results:

• Machine learning models accurately predicted cell sex using sex-dependent gene expression patterns (93%-95% accuracy).
• These models demonstrated high generalizability across datasets.
• Performance surpassed simpler classification models relying only on sex chromosome gene expression (88%-90% accuracy).

Conclusions:

• Sex-dependent gene expression is a viable feature for deconvoluting pooled snRNA-seq data.
• This approach enables cost-effective increases in data throughput and sample size for snRNA-seq studies.
• The study provides a benchmark for machine learning methods applicable to sample deconvolution using inherent biological features.