Deep learning in spatial transcriptomics: Learning from the next next-generation sequencing.

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Spatial transcriptomics (ST) extends single-cell RNA sequencing (scRNAseq) by preserving tissue architecture.
• ST data offers insights into cellular interactions and heterogeneity crucial for understanding complex biological processes.
• Traditional scRNAseq tools and conventional machine learning methods are often inadequate for the high-dimensional, multi-modal nature of ST data.

Purpose of the Study:

• To review existing state-of-the-art computational tools for spatial transcriptomics analysis.
• To explore the emerging role and potential of deep learning (DL) approaches in addressing ST-specific challenges.
• To identify new frontiers and open questions in DL-based ST data analysis.

Main Methods:

• Overview of current ST analysis tools, including those based on traditional statistical and machine learning frameworks.
• In-depth examination of deep learning models applied to ST data challenges such as alignment, spatial reconstruction, and clustering.
• Discussion of the limitations of existing methods and the advantages of DL approaches for ST data.

Main Results:

• Current ST analysis often relies on inadequate traditional methods.
• Deep learning models show promise for improving ST data analysis, with emerging applications in alignment, reconstruction, and clustering.
• The field of DL for ST analysis is nascent but rapidly evolving.

Conclusions:

• Specialized computational tools are essential for robust ST data analysis.
• Deep learning presents a transformative approach for overcoming the complexities and limitations of current ST analysis methods.
• Further research into DL applications is anticipated to drive significant advancements in spatial transcriptomics.