A robust nonlinear low-dimensional manifold for single cell RNA-seq data.

Area of Science:

• Genomics
• Computational Biology
• Biostatistics

Background:

• Single-cell RNA sequencing (scRNA-seq) provides deep insights into cellular heterogeneity.
• Analyzing high-dimensional scRNA-seq data requires dimension reduction techniques.
• Existing methods often require pre-processed data and lack uncertainty estimation.

Purpose of the Study:

• To develop a novel nonlinear latent variable model for analyzing raw, unfiltered scRNA-seq count data.
• To enable uncertainty estimation in the low-dimensional space of scRNA-seq data.
• To improve the analysis of cellular states and heterogeneity.

Main Methods:

• A nonlinear latent variable model based on Gaussian process latent variable model (GPLVM).
• Incorporation of robust, heavy-tailed error modeling using a Student's t-distribution.
• Adaptive kernel learning for estimating nonlinear structures.

Main Results:

• The model effectively estimates low-dimensional nonlinear structures from raw gene counts.
• The approach demonstrates robustness to technical and biological noise.
• Performance was validated across diverse scRNA-seq datasets for downstream tasks like clustering and trajectory inference.

Conclusions:

• The proposed adaptive, robust statistical method is well-suited for raw, high-throughput sequencing data.
• It facilitates visualization, exploration, and uncertainty estimation of cell states.
• This advances the analysis of single-cell gene expression data.