TransST: transfer learning embedded spatial factor modeling of spatial transcriptomics data.

Area of Science:

• Biomedical Research
• Genomics
• Computational Biology

Background:

• Spatial transcriptomics offers insights into tissue biology but faces challenges with low resolution and sequencing depth.
• Extracting reliable biological signals from spatial transcriptomics data remains difficult due to technical limitations.

Purpose of the Study:

• To develop a novel transfer learning framework, TransST, to enhance cell-level heterogeneity inference in spatial transcriptomics data.
• To adaptively leverage external cell-labeled information to overcome data limitations.

Main Methods:

• Proposed a novel transfer learning framework named TransST.
• Applied adaptive leveraging of external cell-labeled information.
• Utilized computational methods to infer cell-level heterogeneity.

Main Results:

• TransST significantly improves existing techniques in both simulation and real-world studies.
• Successfully identified five biologically meaningful cell clusters in a breast cancer study, including distinct cancer subtypes.
• Distinguished between adipose and connective tissues, a capability not matched by other methods.

Conclusions:

• TransST is an effective and robust method for spatial transcriptomics data analysis.
• The framework excels at identifying cell subclusters and their driving biomarkers.
• Demonstrated utility in complex biological samples like breast cancer tissue.