Destin2

Destin2 integrates multimodal single-cell chromatin accessibility data, including scATAC-seq peak accessibility, motif deviation scores, and pseudo-gene activity, to perform dimension reduction, clustering, and trajectory reconstruction for epigenomic analysis.

Key Features:

• Multimodal input: Integrates peak accessibility, motif deviation scores, and pseudo-gene activity from single-cell chromatin datasets.
• Shared manifold learning: Learns a shared manifold representation to combine different types of single-cell chromatin information.
• Dimension reduction: Performs dimension reduction on the integrated manifold for downstream analysis.
• Clustering: Provides clustering of cells based on the integrated chromatin accessibility representation.
• Trajectory reconstruction: Reconstructs cell-state trajectories and transient cellular dynamics from integrated data.
• Benchmarking metrics: Evaluates performance using four distinct metrics with cell-type labels transferred from unmatched single-cell RNA sequencing (scRNA-seq) data.
• Multiomic preservation: Validates preservation of true cell-cell similarities using matched cell pairs in single-cell RNA and ATAC multiomic datasets.

Scientific Applications:

• scATAC-seq analysis: Integrated analysis of scATAC-seq datasets to characterize chromatin accessibility at single-cell resolution.
• Cell differentiation and transient states: Identification and reconstruction of cellular differentiation trajectories and transient cell states.
• Epigenetic regulation studies: Investigation of epigenetic regulatory patterns across individual cells using multiple chromatin-derived features.
• Multiomic integration assessment: Assessment of integration accuracy and preservation of cell-cell relationships in joint single-cell RNA and ATAC analyses.

Methodology:

Destin2 learns a shared manifold from multimodal inputs (peak accessibility, motif deviation scores, pseudo-gene activity), then applies dimension reduction, clustering, and trajectory reconstruction; benchmarking uses four metrics with cell-type labels transferred from unmatched scRNA-seq and matched cell pairs in multiomic datasets as ground truth.