Spatial biology and computational AI for the future

Spatial biology tools have swept through life sciences and pharmaceutical research in recent years. Pathologists, who examine tissue sections for clinical diagnosis, already understand the value of spatial context in tumorigenesis and metastasis. What is new is the ability to deliver in-situ deep molecular profiles from a tissue section.

Consider the knowledge you would typically gain from a histology slide, then add a layer of detail about the expression of dozens of genes or proteins at single-cell and even sub-cellular resolution. That additional layer reveals crucial information about cellular and molecular interactions. Spatial biology has already produced discoveries about the function of immune cells in the tumour microenvironment, the distribution of viral particles in infectious disease, and the progression of neurodegenerative disease. Those discoveries originated in research labs, but pathologists and clinical researchers are already applying them to derive insights from tissue.

Pathology labs: a revolution in progress

Although spatial biology tools remain a research-lab priority, clinical pathology laboratories have already seen significant upstream benefit. The tumour microenvironment (TME) plays a crucial role in cancer development and progression, and characterising it is critical to developing effective therapies. Computer vision and AI applied to TME analysis have enabled the identification of biomarkers with significant prognostic and predictive value — immune cell infiltration, stromal composition and tumour heterogeneity among them — allowing precise identification of the patient subgroups likely to benefit from a specific treatment.

Multiplexing and spatial phenotyping produce unprecedented complexity in tissue-based biomarker expression and cellular organisation. Quantitative digital pathology and AI tools manage that complexity and, in doing so, expand our understanding of disease. Defining and measuring spatial distribution gives systems biology new dimensions. Quantifying the expression and co-expression of multiple biomarkers at cellular level opens a new route to predictive tests in oncology. Spatial biology fills the gap between genomic and transcriptomic analysis on one side and conventional histomorphology evaluation on the other — uncovering the mechanisms behind communication between cancer cells and their microenvironments, which in turn points to new treatment options and new predictive biomarkers.

The challenges ahead

Analysing biopsy slides is among the most promising methods for capturing the complexity of the TME and its function in drug response, because TME information is spatial by nature. Computer vision algorithms could extract millions of data points from whole slide images to describe the tumour and its microenvironment in detail, and to find the subtle immune signatures distinguishing responders from non-responders. But several challenges need addressing first:

• Data analysis. The sheer volume of data generated is the biggest single obstacle. Single-cell sequencing and spatial transcriptomics have given pathologists access to an unprecedented amount of data; analysing and interpreting it is a different matter, requiring computational skill and the right software.
• Integration with existing workflows. These technologies are new, and there is a learning curve for pathologists unfamiliar with them. Integrating spatial data into existing pathology workflows alongside traditional histopathology data requires specialised expertise.
• Standardisation. There is currently no standardisation of spatial biology techniques. Different laboratories use different protocols and methodologies, which produces variability in results and makes comparison across studies difficult.
• Validation. Validating spatial data is challenging given the noise and technical variability that arise during sample processing and analysis.

A holistic solution

Building the tools that serve the next generation of digital pathology and spatial biology is a priority for PRR. Three capabilities matter most:

• Visualising and quantifying the tissue microenvironment — essential to understanding the biological processes occurring in a specific tissue, and to identifying potential therapeutic targets.
• Dynamic computer vision models and image analysis — processing and analysing large volumes of image data to give a fuller view of the cellular and molecular processes within tissue.
• Integrating multi-omics data — combining genomics, transcriptomics, proteomics and metabolomics for a more complete picture of the molecular mechanisms behind disease development and progression.

If you are considering how to bring this into your pathology workflow, talk to us.