scHILL: deciphering individual-level immune cell heterogeneity with single-cell RNA sequencing data.

Area of Science:

• Computational biology
• Immunoinformatics
• Genomics

Background:

• Single-cell RNA sequencing (scRNA-seq) enables high-resolution analysis of cellular heterogeneity.
• Current deep learning methods for scRNA-seq struggle to capture individual-level phenotypic differences.
• Understanding immune cell heterogeneity is crucial for personalized medicine in various diseases.

Purpose of the Study:

• To present scHILL, a novel deep learning framework for deciphering individual-level immune cell heterogeneity from scRNA-seq data.
• To quantify the functional significance of cells and genes at the individual level.
• To improve phenotype prediction and reveal disease-specific immune variations.

Main Methods:

• Integration of a masked autoencoder (MAE) for self-supervised feature learning with a multilayer perceptron (MLP) for individual scoring.
• Pretraining the MAE with data augmentation to handle limited sample sizes effectively.
• Application of the framework to multiple scRNA-seq datasets from infectious, autoimmune, and cancer conditions.

Main Results:

• scHILL demonstrates superior performance in phenotype prediction compared to existing methods.
• The framework successfully reveals individual-level immune cell heterogeneity across diverse disease contexts.
• MAE enables robust feature learning without labels, mitigating challenges associated with small sample sizes.

Conclusions:

• scHILL offers a generalizable approach for interpreting individual scRNA-seq data.
• The framework facilitates the understanding of patient-specific immune responses.
• This work advances the potential for personalized medicine through detailed immune profiling.