PANORAMIC, a new open-source tool, discovered immune signals in colon cancer that were missed by existing methods. It helps researchers distinguish between random cell clusters and meaningful biological patterns of cell arrangements in diseased tissue.
By Jenna Somers
Under a microscope, a slice of tumor tissue looks like a pluralistic community, different types of cells, densely packed in patterns that may suggest coordination and cohesion. For example, immune cells that cluster around tumor cells could indicate that the body is attempting to fight cancer. However, a pattern that may look meaningful could also just be mere happenstance.
Discerning whether a clustering cell pattern is a piece to a much larger disease puzzle or just a random sample of benign tissue has been a persistent challenge for researchers. Analyzing one tissue sample is like analyzing one public opinion poll. It may produce an estimate, but that estimate may come with a much wider margin of error than expected—one that is easily overlooked.
To close this gap, Stanford researchers in the Department of Biomedical Data Science have built a new statistical tool called PANORAMIC (Pooled ANalysis Of VaRiance-Aware Modeling and Inference of Colocalization). The open-source software allows scientists to more precisely measure the spatial composition of different cell types in tissue and to compare those patterns across myriad patients and samples. The team recently published a paper on PANORAMIC in Bioinformatics.
“Most existing methods treat a single tissue measurement as error-free,” said Sylvia Plevritis, the study’s senior author, William H. Hume Professor in the School of Medicine and professor and chair of Biomedical Data Science. “However, every tissue sample is just a snapshot of a bigger picture. Each measurement carries some degree of uncertainty that, if left unaccounted, can obscure important biological differences between patients, or worse, cause you to mistake the noise for a meaningful discovery.”
How it works
The development of PANORAMIC was led by Jake Chang, a graduate student in Biomedical Data Science, co-advised by Professors Plevritis and Lu Tian.
PANORAMIC works in three steps. First, it measures how different types of cells cluster around each other within a single tissue sample–in other words, the co-localization of two cell types. Next, it estimates the uncertainty of the co-localization measures by repeatedly resampling different small patches of the tissue sample. This process indicates whether a specific co-localization pattern is consistent across the entire tissue sample or driven by localized region. Lastly, it uses a meta-analysis technique to appropriately adjust the uncertainty across samples, patients and conditions in comparing the co-localization pattern between two groups.
Through this process, PANORAMIC offers researchers an estimate of how clustered cells are and a degree of confidence in that estimate.
PANORAMIC proves itself
To test PANORAMIC, the researchers applied it to tissue samples from colorectal cancer patients. They compared two types of tumors: Crohn’s-like tumors and those with diffuse inflammatory infiltration. Within the Crohn’s-like tumors, they found that immune system cells were significantly more tightly clustered together, a pattern consistent with the organized immune structures known to be associated with better patient outcomes.
Notably, standard statistical methods missed these findings, demonstrating the importance of PANORAMIC’s more thorough method. Moreover, the team successfully repeated the analysis on a dataset of head and neck cancer samples, showing that the tool can be broadly applied across different types of cancer.
Why it matters
New imaging technologies make it possible to map thousands of different cells and their precise locations within a tissue sample, and spatial biology is one of the fastest growing domains of cancer and disease research. Unfortunately, the statistical methods for analyzing this data haven’t kept up with the complexity of the data itself.
PANORAMIC offers a promising solution and a more rigorous method for understanding patterns within a single sample and across patient groups, as well as whether a spatial pattern reveals disease mechanism or is just random noise. These distinctions are important for the practical goal of turning spatial patterns in tissue into reliable biomarkers that doctors could reference to help predict which patients will respond best to certain treatments and to understand why some tumors are more successful than others at evading the immune system.
PANORAMIC is freely available as an open-source R software package, allowing any lab working with spatial tissue imaging data to apply it to their own research.