Cell Decoder: decoding cell identity with multi-scale explainable deep learning.

Area of Science:

• Single-cell omics
• Computational biology
• Machine learning for biology

Background:

• Understanding cell diversity and function is crucial.
• Single-cell omics data offers insights but lacks interpretability.
• Current deep learning models struggle with multi-scale cell representation.

Purpose of the Study:

• To develop a novel deep learning model for multi-scale cell representation.
• To integrate biological prior knowledge for enhanced interpretability.
• To improve cell identity decoding and data integration.

Main Methods:

• Introduction of Cell Decoder, a novel deep learning model.
• Integration of biological prior knowledge into the model architecture.
• Application of automated machine learning and post hoc analysis.

Main Results:

• Cell Decoder provides multi-scale cell representations.
• The model decodes cell identity with superior performance.
• Cell Decoder offers multi-view interpretability and facilitates data integration.

Conclusions:

• Cell Decoder reveals multi-scale heterogeneity in human bone and mouse embryonic data.
• The model provides a powerful framework for studying cellular diversity.
• Enhanced understanding of cell identity and function is achieved.