Decoding Complexities: The Critical Role of Deconvolution in Spatial Transcriptomics

Spatial transcriptomics is revolutionizing our understanding of cellular environments by merging high-throughput gene expression data with precise spatial localization within tissues. It stands at the forefront of genomic research, offering a transformative approach, unlike traditional transcriptomics analyzes gene expression in bulk tissue samples or isolated cells without any spatial context.

The integration of sequencing with spatial understanding allows scientists to map the expression of thousands of genes directly onto tissue sections, thereby painting a detailed picture of the cellular function and interaction, which is unachievable by traditional bulk and single cell RNA sequencing methods. This technique is crucial for exploring tissue architecture, cellular interactions, and the heterogeneity within biological systems, providing invaluable insights into developmental biology, disease pathology, and therapeutic targeting.

Importance of Deconvolution in Spatial Transcriptomics

Various spatial transcriptomics (ST) technologies have been developed and utilized across diverse tissues such as mouse and human brains, lymph nodes, and the heart, offering new insights into cellular communication networks in different contexts.

However, sequencing-based ST methods like the 10x Genomics Visium platform and Slide-seq are structured as spot-by-gene matrices and require supplementary data for cellular identification.

Conversely, imaging-based ST technologies such as seqFISH+, MERFISH, 10x Genomics Xenium, and NanoString CosMx offer subcellular resolution but have limited gene throughput, detecting only hundreds of custom genes, which restricts their discovery potential compared to whole transcriptome-wide spatial technologies. Consequently, the integration of whole transcriptome-wide ST data with matched single-cell RNA sequencing (scRNA-seq) data holds significant importance for advancing biological discoveries.

What is Deconvolution in Spatial Transcriptomics?

Deconvolution is a sophisticated computational strategy employed in spatial transcriptomics to dissect complex data obtained from tissue samples. Each 'spot' on a tissue captured in spatial datasets can contain transcripts from multiple cell types, making it a composite signal of mixed gene expression. Deconvolution works by untangling these mixed signals to attribute specific gene expressions to distinct cell types present within each spot. Using reference datasets from single-cell RNA sequencing, where individual cell types are already annotated, deconvolution tools like RCTD (Robust Cell-Type Decomposition) help predict the cell-type composition in each spot. This process is critical for accurately understanding the cellular architecture within tissues, which are essential for deciphering biological functions and disease mechanisms.

Methods of Deconvolution in Spatial Transcriptomics

Traditional bulk transcriptome analysis has played a vital role in elucidating the molecular mechanisms underlying complex biological processes, yet it has struggled to fully reveal the inherent diversity within samples. Bulk RNA-seq data captures the combined gene expression levels of all cells in a sample, limiting its ability to explore cellular heterogeneity. This can obscure important analyses such as differential gene expression due to variations in cell type proportions.

The swift advancement of spatial transcriptomics methodologies has unveiled fresh perspectives on comprehending tissue organization and operations.

Yet, the gene expressions recorded at a specific spot may amalgamate inputs from multiple cells, given the limited resolution of current ST technologies.

To tackle these challenges associated with Spatial Transcriptomics data, numerous computational approaches have been devised to unravel the spatial mixtures of each ST spot into individual cell types, often leveraging single-cell RNA-seq data. For instance, enrichment-based methods like Seurat and MIA assess the significance score or probability of different cell types' presence in each spot. Conversely, other deconvolution techniques aim to estimate the proportion of cell types at each spatial location using diverse strategies such as linear regression models (e.g., SPOTlight, spatialDWLS), probabilistic models (e.g., RCTD, cell2location), or deep learning methods (e.g., DSTG). Furthermore, a handful of reference-free methods (e.g., STdeconvolve) have been proposed, which deconvolve ST data independently of scRNA-seq data.

Spatial transcriptomics deconvolution relies on cell-type-specific gene expression data and specialized spatial techniques. This information is commonly sourced from single-cell RNA-sequencing (scRNA-seq) studies, traditionally utilized for deconvolution in bulk RNA-seq data through methods like MuSiC, SCDC, and Bisque. While these methods can be adapted for spatial transcriptomics, newer approaches discussed above are being tailored specifically for this purpose. The common steps usually include:

• Creation of a Reference Object: A matrix consisting of gene expression data from identified single cells, labeled by cell type is created.
• Query Object Setup: The spatial transcriptomics data, formatted similarly to the reference but representing the mixed cell types within tissue spots.
• Comparison and Attribution: Algorithms compare the query data against the reference, predicting cell type compositions for each spot based on similarities in gene expression patterns.

This method is essential for enhancing the resolution of spatial transcriptomic analyses, allowing researchers to observe how cells of different types are organized spatially and how they interact within the tissue microenvironment.

Challenges in Deconvolution of Spatial Transcriptomics Data

Deconvolution in spatial transcriptomics is a powerful technique to gain biological insights but also faces several significant challenges:

• Reference Data Quality: The accuracy of cell type attribution heavily depends on the quality and comprehensiveness of the single-cell dataset being used as a reference. Poorly annotated or incomplete reference data can lead to incorrect cell type attribution.
• Computational Demands: Handling and analyzing large datasets require substantial computational power and efficient algorithms to manage the complexity.
• Data Integration and Scaling: Integrating spatial data with traditional transcriptomic data can introduce scale and normalization challenges that affect the analysis.

Mitigating Challenges with Polly at Elucidata

Elucidata’s data & AI-cloud- Polly, provides advanced solutions to address the challenges associated with spatial transcriptomics data analysis. Polly addresses these challenges through a suite of tools designed to streamline and enhance the analysis of spatial transcriptomics data:

1. Data Harmonization

Polly achieves data consistency and comparability by integrating spatial and gene expression data from multiple sources. It includes an extensive library of scRNA-seq datasets, which researchers can utilize to select reference datasets that closely match the cell type composition expected in their spatial studies.  It seamlessly combines Spatial Transcriptomics (SRT) data, including raw counts matrices, spatial coordinates, imaging data, and metadata from diverse public and private collections. This integration facilitates a more nuanced understanding of biological complexities.

2. Enhanced Quality Control

Polly conducts extensive quality control measures, including approximately 50 QA/QC checks that cover metadata quality, normalization, batch effect correction, and measurement accuracy. This thorough quality assurance helps ensure the data’s integrity and suitability for detailed analyses. Applying these rigorous QC procedures developed by experts to both spatial and scRNA-seq reference data, enhance the performance of data deconvolution.

3. Customizable Pipelines

Polly allows users to customize processing pipelines according to their specific research needs, enhancing the accuracy of data deconvolution. It also offers 30+ ready-to-use, ETL processing pipelines for users. These pipelines help maintain consistency across datasets, ensuring researchers can replicate studies based on original raw counts.

4. Advanced Computational Resources

‍Polly is equipped with powerful computational resources that efficiently manage large datasets, significantly reducing the computational load on researchers. It features a unified data architecture and API-driven access to high-quality, harmonized data, simplifying the integration and querying processes.

5. Interactive Visualization Tools

The platform includes advanced visualization tools that help to explore and interpret spatial data. Integrated tools like CellxGene VIP allow for on-the-fly analysis and visualization of spatial transcriptomics data, aiding in the mapping of cell-type distributions and tissue interactions. These tools can incorporate additional contextual information, such as annotated tissue regions, enhancing the depth and utility of spatial dataset analyses.

Read More

The deconvolution of spatial transcriptomics data is crucial for advancing our understanding of cellular and molecular biology within the spatial contexts of tissues. By resolving the complexities associated with mixed cell type signals, researchers can gain deeper insights into the cellular dynamics that define health and disease. Solutions provided by Elucidata’s represents pivotal advancement in this field, offering powerful tools that not only tackle the inherent challenges of deconvolution but also enhance the overall efficiency and effectiveness of spatial transcriptomics research. As the demand for detailed spatial biological analysis grows, platforms like Polly are essential for driving discoveries and innovations in biomedicine.

To learn more about how Polly can transform your research, connect with us or reach out to us at [email protected].

‍