Scalable inference of transcriptional variability with BASiCS

Area of Science:

• Computational Biology
• Bioinformatics
• Statistical Genetics

Background:

• Single-cell RNA sequencing (scRNA-seq) generates high-dimensional data requiring robust analytical methods.
• Existing Bayesian models for scRNA-seq data analysis face scalability challenges with increasing dataset sizes.
• Accurate normalization and noise quantification are crucial for reliable interpretation of scRNA-seq data.

Purpose of the Study:

• To extend the BASiCS (Bayesian Analysis of Single-Cell Sequencing data) framework for scalable Bayesian inference.
• To develop and evaluate a novel divide and conquer inference scheme for large-scale scRNA-seq datasets.
• To compare the performance of the new approach against standard Markov Chain Monte Carlo (MCMC) and Approximate Dual Variational Inference (ADVI) methods.

Main Methods:

• Implementation of a divide and conquer Bayesian hierarchical model for scRNA-seq data.
• Simultaneous data normalization and technical noise quantification within the BASiCS framework.
• Comparative performance analysis using accuracy and scalability metrics against MCMC and ADVI.

Main Results:

• The divide and conquer approach significantly improves scalability for large scRNA-seq datasets.
• Accurate and efficient Bayesian inference is achieved while preserving the interpretability of the BASiCS model.
• The extended BASiCS framework demonstrates superior performance in handling complex single-cell data.

Conclusions:

• The divide and conquer inference scheme provides a scalable solution for Bayesian analysis of large scRNA-seq datasets.
• This advancement facilitates more comprehensive and reliable analysis of gene expression variability and cell population heterogeneity.
• The enhanced BASiCS framework maintains flexibility and interpretability, crucial for biological discovery in single-cell genomics.