CLAIRE

CLAIRE integrates single-cell RNA sequencing (scRNA-seq) datasets using contrastive learning to correct batch effects while preserving cellular heterogeneity for accurate comparative analyses.

Key Features:

• Contrastive Learning Framework: Employs a contrastive learning-based approach to learn representations that align shared cell types across batches and correct batch effects in scRNA-seq integration.
• Mix-Heterogeneity Trade-off: Explicitly optimizes the trade-off between batch mixing and preservation of cellular heterogeneity rather than relying on ad hoc balancing.
• Construction Strategy: Dynamically generates positive pairs by augmenting inter-batch mutual nearest neighbors (MNN) with intra-batch k-nearest neighbors (KNN) to enhance coverage of shared cell-type distributions across batches.
• Refinement Strategy: Leverages the memory effect of deep neural networks to automatically filter potential false positive pairs produced during construction.

Scientific Applications:

• scRNA-seq dataset integration: Integrating multiple scRNA-seq datasets to correct batch effects while retaining biological heterogeneity for downstream analysis.
• Comparative analyses across studies and conditions: Facilitating comparative studies across experiments or conditions to investigate cellular diversity and function.

Methodology:

Two complementary computational strategies for positive-pair selection in contrastive learning: construction that augments inter-batch MNN with intra-batch KNN to broaden coverage, and refinement that uses the memory effect of deep neural networks to filter false positive pairs.