Model-based dimensionality reduction for single-cell RNA-seq using generalized bilinear models.

Area of Science:

• Computational biology
• Genomics
• Bioinformatics

Background:

• Dimensionality reduction is crucial for single-cell RNA-seq (scRNA-seq) data analysis.
• Standard methods like PCA can introduce artifacts and obscure true biological signals.
• Existing count-based models are often computationally intensive and lack uncertainty quantification.

Purpose of the Study:

• To develop a novel, scalable, and model-based dimensionality reduction method for scRNA-seq data.
• To address limitations of existing methods in handling large datasets and quantifying uncertainty.
• To improve the biological interpretability of low-dimensional embeddings.

Main Methods:

• Developed scGBM, a method utilizing a Poisson bilinear model for scRNA-seq dimensionality reduction.
• Implemented a fast estimation algorithm based on iteratively reweighted singular value decompositions.
• Incorporated uncertainty quantification for cell latent positions and clustering confidence assessment.

Main Results:

• scGBM scales effectively to datasets with millions of cells.
• The method produces low-dimensional embeddings that better capture biological information.
• scGBM successfully removes unwanted variation from scRNA-seq data.
• Uncertainty quantification aids in assessing cell clustering confidence.

Conclusions:

• scGBM offers a computationally efficient and statistically robust approach to scRNA-seq dimensionality reduction.
• The method enhances the identification of true biological variability and reduces spurious heterogeneity.
• scGBM provides a valuable tool for analyzing large-scale scRNA-seq datasets and improving downstream analyses like cell clustering.