Representation learning for multi-modal spatially resolved transcriptomics data

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Spatial transcriptomics offers detailed molecular and morphological insights while preserving spatial context.
• Integrating multi-modal spatial transcriptomics data remains a challenge for precision medicine.

Purpose of the Study:

• To develop a novel deep learning model for joint integration of spatial, transcriptomics, and morphology data.
• To improve the accuracy of spot representations and cell clustering in spatial transcriptomics analysis.

Main Methods:

• Introduced AESTETIK, a convolutional deep learning model for multi-modal data integration.
• Applied AESTETIK to various datasets, including structured (brain) and heterogeneous (cancer) tissues.
• Utilized widely adopted technology platforms like 10x Genomics™ and NanoString™.

Main Results:

• AESTETIK significantly improved cluster assignments across multiple datasets and platforms.
• Achieved a 21% increase in median ARI for structured tissues (e.g., brain) compared to state-of-the-art methods.
• Demonstrated superior performance on cancer tissues, with notable improvements in breast cancer (2-fold), melanoma (79%), and liver cancer (21%).

Conclusions:

• AESTETIK enables accurate spot representation by integrating spatial, transcriptomics, and morphology data.
• The model enhances cell clustering performance, particularly in complex tissue types.
• These advances facilitate a multi-modal understanding of biological processes and support precision medicine applications.