scPair: Boosting single cell multimodal analysis by leveraging implicit feature selection and single cell atlases.

Area of Science:

• Genomics
• Computational Biology
• Single-cell Analysis

Background:

• Multimodal single-cell assays provide rich cellular information by profiling multiple features (e.g., chromatin, mRNA) within the same cell.
• High dimensionality and shallow sequencing depth in multimodal data present significant analytical challenges.
• Existing methods struggle to efficiently integrate and interpret diverse single-cell data types.

Purpose of the Study:

• To introduce scPair, a novel computational framework for analyzing multimodal single-cell data.
• To address challenges posed by high dimensionality and shallow sequencing depth in multimodal single-cell assays.
• To improve the accuracy and efficiency of cell state mapping and gene regulatory element identification across modalities.

Main Methods:

• scPair utilizes dual encoder-decoder structures trained on paired multimodal single-cell data.
• An implicit feature selection approach is employed to handle high-dimensional data.
• The framework aligns cell states and predicts features between different modalities (e.g., chromatin to mRNA).

Main Results:

• scPair demonstrates superior accuracy and execution time compared to existing multimodal single-cell analysis methods.
• The framework effectively facilitates downstream analyses, including trajectory inference.
• scPair successfully augments smaller multimodal datasets with larger unimodal atlases to enhance statistical power for identifying active transcription factors.

Conclusions:

• scPair provides an effective solution for analyzing complex multimodal single-cell data.
• The framework enhances the ability to map cell states and link regulatory elements to gene expression.
• scPair has the potential to significantly advance research in developmental biology and disease by improving the analysis of single-cell atlases.