scBFA

scBFA models gene detection patterns to mitigate technical variation and produce low-dimensional embeddings for downstream analysis of scRNA-seq and scATAC-seq datasets.

Key Features:

• Binary Factor Analysis Model: Implements a binary factor analysis model that models gene detection patterns to reduce dimensionality and mitigate noisy expression profiles.
• Detection-based Modeling: Focuses on gene detection patterns rather than feature quantification, which is advantageous when detection noise is lower than quantification noise.
• Covariate Matrices Integration: Accepts cell-level and gene-level covariate matrices (X and Q) to adjust nuisance variance such as batch effects.
• Low-Dimensional Embedding Output: Produces a low-dimensional embedding matrix that captures biological signal while minimizing technical noise for downstream analyses.
• Performance in Downstream Tasks: Demonstrates state-of-the-art performance in cell type classification and trajectory inference tasks.

Scientific Applications:

• Cell Type Identification: Enhances accuracy of identifying distinct cell types in complex single-cell datasets, including scRNA-seq and scATAC-seq.
• Trajectory Inference: Improves trajectory inference by reducing technical noise and clarifying developmental and differentiation pathways.

Methodology:

Applies a binary factor analysis model to gene detection patterns and incorporates cell-level (X) and gene-level (Q) covariate matrices to produce a low-dimensional embedding.