BaySiCle

BaySiCle performs Bayesian inference–based imputation of missing values in single-cell RNA sequencing (scRNA-seq) datasets to account for technical variability and produce biologically plausible transcript distributions.

Key Features:

• Bayesian inference: Uses Bayesian statistics to model uncertainty associated with missing scRNA-seq observations for more robust imputation.
• k-Nearest Neighbors (kNN): Employs k-nearest neighbors (kNN) to identify similar cells within a batch and inform imputation using local neighborhoods.
• Batch-effect utilization: Leverages batch effects—systematic differences between experimental batches—by using within-batch information under the assumption that cells in a batch originate from the same tissue.
• Prior estimation from within-batch expression: Derives priors from expression values across cells within the same batch to contextualize Bayesian inference.
• Local-effect modeling: Models local effects by focusing on immediate neighborhood characteristics to make imputation sensitive to cell-specific local structure.

Scientific Applications:

• scRNA-seq imputation: Imputes missing observations in single-cell RNA sequencing (scRNA-seq) datasets to mitigate technical variability during downstream analysis.
• Transcript distribution generation: Generates realistic transcript distributions that align with single molecule fluorescence in situ hybridization (smFISH) measurements.
• Evaluation on real and simulated data: Demonstrated on sample scRNA-seq datasets and simulated expression data to assess imputation plausibility.

Methodology:

BaySiCle applies Bayesian inference integrated with k-nearest neighbors (kNN), using priors estimated from expression values across cells within the same batch and leveraging batch and local neighborhood effects to impute missing scRNA-seq values.