Shamini Ayyadhury, PhD

Postdoctoral Fellow (University of Toronto, Princess Margaret Cancer Research Center)

Director, Panoramics – A Vision

Learning Objectives

• By the end of this lecture, you will:

​

• Think critically about spatial biology and analysis
• Develop an intuition on the factors that affect or influence spatial analysis
• Develop foundational skills in spatial analysis

Why do we want to cluster?

• Why do we want to cluster?
• We have to critically evaluate why do we want spatial?
• Spatial means we want to understand
• Networks
• Interactions
• Communication
• PATTERNS that are beyond linear
• Non-spatial versus spatial
• Combination with other computational methods
• Segmentation
• Bin or scale size
• Kernel-based smoothing
• Interpretation of gene modules of DE
• Think critically of tissue type and tissue size
• Some tissues have patterns that do not necessitate the effort for spatial analysis
• i.e when single cell labels identify domains
• If the scale of pattern recognition is is bigger than the limit of the computational spatial tool applied

Why do we want to cluster?

• Why do we want to cluster?

​

Ayyadhury, S, Unpublished data, Xenium analysis of GBM

We have to critically evaluate why do we want spatial?

Inferring histology-associated gene expression gradients in spatial transcriptomic studies, Nature Communications volume 15, Article number: 7280 (2024)

• Spatial means we want to understand
• Networks
• Interactions
• Communication
• PATTERNS that are beyond linear

​

​

​

​

But many of these do not require spatial clustering

Many of these interactions require granular analysis

• Spatial means we want to understand
• Networks
• Interactions
• Communication
• PATTERNS that are beyond linear

​

​

​

​

But many of these do not require spatial clustering

Many of these interactions require granular analysis

SPATIAL IS THE UNDERSTANDING OF COLLECTIVE MECHANISMS OR IDENTIFYING DOMAINS

Non-spatial versus spatial

BANKSY unifies cell typing and tissue domain segmentation for scalable spatial omics data analysis, Nature Genetics volume 56, pages431–441 (2024)

Non-spatial versus spatial

BANKSY unifies cell typing and tissue domain segmentation for scalable spatial omics data analysis, Nature Genetics volume 56, pages431–441 (2024)

Non-spatial versus spatial

BANKSY unifies cell typing and tissue domain segmentation for scalable spatial omics data analysis, Nature Genetics volume 56, pages431–441 (2024)

?

Non-spatial versus spatial

BANKSY unifies cell typing and tissue domain segmentation for scalable spatial omics data analysis, Nature Genetics volume 56, pages431–441 (2024)

?

Applied a smoothing Kernel

Combination with other computational methods

• Segmentation
• Bin or scale size
• Kernel-based smoothing

• Combination with other computational methods
• Segmentation
• Bin or scale size
• Kernel-based smoothing

Segmentation is not a pre-requisite for spatial clustering

FICTURE: scalable segmentation-free analysis of submicron-resolution spatial transcriptomics, Nature Methods volume 21, pages1843–1854 (2024�

Interpretation of gene modules of DE

Inferring histology-associated gene expression gradients in spatial transcriptomic studies, Nature Communications volume 15, Article number: 7280 (2024)

BANKSY unifies cell typing and tissue domain segmentation for scalable spatial omics data analysis, Nature Genetics volume 56, pages431–441 (2024)

Interpretation of gene modules of DE

Inferring histology-associated gene expression gradients in spatial transcriptomic studies, Nature Communications volume 15, Article number: 7280 (2024)

BANKSY unifies cell typing and tissue domain segmentation for scalable spatial omics data analysis, Nature Genetics volume 56, pages431–441 (2024)

Depends on the question/method

Example 1 below

1. Defined a trajectory across spatial manually annotated spatial clusters
2. Mathematics and spatial analysis were a derivative of that clinical aspect (GBM)

​

Example 2

1. Banksy’s main objective is to identify domains
2. The tri-matrix output does not offer much in terms of DE

Think critically of tissue type and tissue size

• Some tissues have patterns that do not necessitate the effort for spatial analysis
• If the scale of pattern recognition is is bigger than the limit of the computational spatial tool applied

Think critically of tissue type and tissue size

• Some tissues have patterns that do not necessitate the effort for spatial analysis
• If the scale of pattern recognition is is bigger than the limit of the computational spatial tool applied

Think critically of tissue type and tissue size

• Some tissues have patterns that do not necessitate the effort for spatial analysis
• If the scale of pattern recognition is is bigger than the limit of the computational spatial tool applied

Think critically of tissue type and tissue size

• Some tissues have patterns that do not necessitate the effort for spatial analysis
• If the scale of pattern recognition is is bigger than the limit of the computational spatial tool applied

Think critically of tissue type and tissue size

• Some tissues have patterns that do not necessitate the effort for spatial analysis
• If the scale of pattern recognition is is bigger than the limit of the computational spatial tool applied

Integrative spatial analysis reveals a multi-layered organization of glioblastoma, Cell, Volume 187, Issue 10P2485-2501.E26May 09, 2024

Think critically of tissue type and tissue size

• Some tissues have patterns that do not necessitate the effort for spatial analysis
• If the scale of pattern recognition is is bigger than the limit of the computational spatial tool applied

Go through example papers and materials & methods

Practical session

Module3a script - Take you through the process of

1. Dimensional reduction

2. non-spatial clustering

3. Spatial clustering(Banksy)

Module3b script – You will perform Spatial Clustering (blanked out script)

1. Savannah – filtered object
2. In this module, you will be using unfiltered matrix to re-create another Seurat object in module3a script.
3. This script is blanked out – so you will be given time to fill in
4. But we will review the code and we will compare the filtered with the unfiltered
5. This part is to allow you to understand how pre-processing steps can affect domain analysis

References

1. Hu, Y., Xie, M., Li, Y., Rao, M., Shen, W., Luo, C., Qin, H., Baek, J., & Zhou, X. M. (2024). Benchmarking clustering, alignment, and integration methods for spatial transcriptomics. Genome Biology, 25, 212. https://doi.org/10.1186/s13059-024-03361-0

2. Inferring histology-associated gene expression gradients in spatial transcriptomic studies. (2024). Nature Communications, 15, 7280. https://doi.org/xxxxx (Replace with correct DOI)

3. BANKSY unifies cell typing and tissue domain segmentation for scalable spatial omics data analysis. (2024). Nature Genetics, 56, 431–441. https://doi.org/xxxxx (Replace with correct DOI)

4. FICTURE: Scalable segmentation-free analysis of submicron-resolution spatial transcriptomics. (2024). Nature Methods, 21, 1843–1854. https://doi.org/xxxxx (Replace with correct DOI)

​

Contact information :

Shamini Ayyadhury

​

​

Academic brainstorming & consultation : shamini.ayyadhury@utoronto.ca

Panoramics – A Vision : shamini.ayyadhury@panoramics-a-vision.com

​

�

Connect with me

We are on a Coffee Break & Networking Session

Workshop Sponsors: