DSCT: a novel deep-learning framework for rapid and accurate spatial transcriptomic cell typing.

Area of Science:

• Neuroscience
• Computational Biology
• Genomics

Background:

• Understanding brain cell composition and gene expression at spatial resolution is key to deciphering neural functions.
• Spatial transcriptomics offers unprecedented insights into cellular neighborhoods but requires advanced analytical tools.

Purpose of the Study:

• To introduce Deep Neural Network-based Spatial Cell Typing (DSCT), an innovative computational framework for high-resolution spatial cell typing.
• To provide a rapid, accurate, and computationally efficient method for analyzing spatial transcriptomic data.

Main Methods:

• Developed DSCT, integrating an enhanced gene-selection strategy with a lightweight deep neural network for training.
• Applied DSCT to diverse spatial transcriptomic datasets from various brain regions, species, and platforms.

Main Results:

• DSCT demonstrated exceptional accuracy in identifying cell types, including finer subtypes, across different datasets.
• The framework exhibited high efficiency and remarkable processing speed with reduced computational demands.
• DSCT proved versatile and adaptable to various spatial transcriptomic data types.

Conclusions:

• DSCT offers a powerful and efficient solution for spatial cell typing in complex biological systems.
• This novel approach facilitates deeper exploration of spatial cell-type organization and gene expression in the nervous system.
• DSCT advances the understanding of biological functions and neurological pathologies through precise spatial analysis.