Double-jeopardy: scRNA-seq doublet/multiplet detection using multi-omic profiling

Area of Science:

• Genomics and Computational Biology
• Single-cell analysis techniques
• Machine learning in bioinformatics

Background:

• Accurate identification of biological signals in single-cell RNA sequencing (scRNA-seq) data requires the computational detection and exclusion of cellular doublets and multiplets.
• Existing methods often lack sensitivity in identifying both heterotypic and homotypic doublets/multiplets.
• Multi-omic single-cell data offers potential for improving data quality.

Purpose of the Study:

• To develop and validate a machine learning approach for sensitive and specific doublet/multiplet detection in scRNA-seq data.
• To leverage VDJ-seq and CITE-seq data for improved doublet/multiplet identification.
• To enhance the generation of high-quality scRNA-seq datasets.

Main Methods:

• A machine learning model was developed to predict the presence of doublets/multiplets.
• The model utilizes transcriptional features from hybrid droplets identified using VDJ-seq and/or CITE-seq data.
• The method was evaluated on scRNA-seq samples, particularly those rich in inflammatory cells.

Main Results:

• The proposed machine learning approach demonstrates high sensitivity and specificity in doublet/multiplet detection.
• The method effectively utilizes multi-omic single-cell information (VDJ-seq, CITE-seq) for accurate predictions.
• Validation in inflammatory-cell-dominant samples confirms the method's robustness.

Conclusions:

• This machine learning strategy provides a powerful tool for ensuring high-quality scRNA-seq data by accurately identifying and excluding doublets/multiplets.
• Leveraging multi-omic data significantly improves the reliability of single-cell analyses.
• The method is particularly valuable for complex biological samples like those dominated by inflammatory cells.