All Eyes on IRIS

Physical exams, lab tests, medical imaging studies, biopsies, and genetic testing come together to provide a relatively comprehensive characterization of an individual’s health. But what if we could also integrate modern functional and spatial genomics – areas that have seen major advances in the past decade.

Researchers at the University of Michigan have recently developed a new AI method that analyzes spatial transcriptomics data from tissue samples (1). To learn more about “integrative and reference-informed tissue segmentation” – shortened to IRIS – and its ability to transform our understanding of diseases, we spoke with lead author Ying Ma, former PhD Student at the University of Michigan and current Assistant Professor at Brown University, and senior author Xiang Zhou, Professor at the University of Michigan, USA.

Please tell us about IRIS…
 

The IRIS method is a groundbreaking computational AI approach that enables biologists to investigate the complex architecture of tissues – uncovering dynamic processes that govern tissue structure during both normal development and disease progression.

In its essence, IRIS is a machine learning method that uses reference data to identify spatial domains in tissue slices from spatial transcriptomics. It builds on a scalable penalized regression AI model and works with various spatial resolutions, from spot-level to single-cell and subcellular levels. IRIS integrates single-cell RNA sequencing data to accurately determine cell types within spatial domains. It precisely detects spatial domains and uses cell type similarities to map tissue structures in both healthy and diseased states. This helps identify molecular, cellular, and tissue changes during disease progression, improving our understanding of diseases and aiding in diagnostics and therapy development.

How could IRIS support tissue and tumor analysis?
 

We’ve demonstrated the advantages of IRIS through in-depth analysis of six spatial transcriptomics (SRT) datasets that encompass a variety of technologies, tissues, species, and resolutions. This technology can work with high capacity databases and performs at high speeds with accuracy – revealing intricate brain structures, uncovering heterogeneity within the tumor microenvironment, and detecting structural changes in diabetes-affected testis. IRIS opens the door for specifically mapping the diverse cellular environments within tumors and providing crucial insights into tumor growth, metastasis, and treatment resistance. 

By providing a more nuanced understanding of tissue and tumor microenvironments, IRIS has the potential to significantly influence clinical diagnostics and therapeutic intervention.

We could also envisage IRIS being used in developmental biology to integrate multiple tissue slides at different developmental stages and time points – allowing for a comprehensive characterization of the complex processes of tissue development, organ formation, and disease progression. 

There are also opportunities for this technology to crossover into neurodegenerative research and further clinical diagnostic lab work.

Can IRIS be further improved?
 

We’re looking at enhancing IRIS’s ability to account for cell type compositional similarity across tissue slide locations by incorporating diverse spatial correlation patterns through multiple kernels, such as Gaussian and periodic. 

Though this version of IRIS relies on k-means clustering for stability and scalability, future iterations of IRIS could use Louvian or Leiden clustering algorithms, which could offer many unique advantages. Further testing is required to adjust for potential impacts. 

With advances in image segmentation and cell alignment techniques, we’re hoping to integrate histological image collection capabilities alongside spatial transcriptomics data. Furthermore, we aim to expand IRIS’ capabilities to include mapping single cells from scRNA-seq onto spatial locations in SRT, broadening the scope of its applicability in spatial genomics and clinical research.

What role do you think AI will play in the future of disease diagnosis?
 

The integration of AI into clinical settings will likely fundamentally transform the future of disease diagnosis – offering unmatched precision and efficiency. AI’s ability to process vast and intricate datasets can enable earlier and more accurate detection of diseases, which is crucial for personalized treatment moving forward. 

Looking ahead, the use of spatial omics data on individual patients could allow AI to construct detailed patient atlases – enabling the formulation of tailored treatment plans and diagnoses that are finely adjusted to each patient’s unique biological and genomic makeup.