Cell-type annotation with accurate unseen cell-type identification using multiple references

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) enables detailed tissue cellular composition analysis.
• Automated cell-type annotation relies on comprehensive reference datasets, which often lack cell types present in query data.
• Identifying novel or previously unseen cell types is crucial for accurate annotation and biological insights.

Purpose of the Study:

• To develop a novel computational method, mtANN, for automated scRNA-seq data annotation.
• To accurately identify previously unseen cell types within query datasets using multiple references.
• To enhance the accuracy and scope of cell-type annotation in complex biological samples.

Main Methods:

• Proposed mtANN (multiple-reference-based scRNA-seq data annotation) method.
• Integrated deep learning and ensemble learning for improved prediction.
• Introduced a novel metric considering three aspects to differentiate unseen from shared cell types.
• Developed an adaptive thresholding method for unseen cell-type identification.

Main Results:

• mtANN demonstrated superior performance in identifying unseen cell types compared to state-of-the-art methods.
• The method achieved high accuracy in cell-type annotation across benchmark datasets.
• Validated predictive power on COVID-19 related scRNA-seq datasets.
• Source code and tutorial are publicly available.

Conclusions:

• mtANN effectively addresses the challenge of unseen cell types in scRNA-seq data annotation.
• The method enhances biological discovery by revealing novel cellular populations.
• mtANN offers a robust and accurate solution for scRNA-seq data analysis.