Quantifying pathological progression from single-cell transcriptomic data with scPSS.

Area of Science:

• Computational Biology
• Genomics
• Biostatistics

Background:

• Single-cell data analysis is rapidly advancing, enabling comparisons across conditions.
• A critical gap exists in quantifying pathological changes at the single-cell level from healthy states.
• Existing methods lack the ability to measure cell-level pathological progression in a semisupervised manner.

Purpose of the Study:

• To introduce single-cell Pathological Shift Scoring (scPSS), a novel statistical method for quantifying pathological shifts in individual cells.
• To provide a statistically rigorous measure of how much a query cell deviates from a healthy reference population.
• To enable pathological progression assessment without requiring labeled diseased data for training.

Main Methods:

• scPSS calculates pathological shift scores based on the distance of a query cell to its k-th nearest reference cell in principal component space.
• Euclidean distances are computed in the top n principal component space of gene expression data.
• A null model is established using the distribution of shift scores from reference healthy cells, enabling P-value assignment.

Main Results:

• scPSS provides a statistically rigorous and interpretable measure of pathological deviation for individual cells.
• The method demonstrates superior accuracy and efficiency compared to adapted state-of-the-art supervised models in benchmarking.
• Aggregated cell-level scPSS scores can effectively predict health conditions at the individual level.

Conclusions:

• scPSS offers a powerful, semisupervised approach to quantify cell-level pathological shifts, addressing a key gap in single-cell analysis.
• The method's statistical rigor and efficiency make it a valuable tool for disease research.
• scPSS has potential applications in disease diagnosis and monitoring by enabling individual-level health predictions.