Visualising gene expression provides us with a glimpse into the universe of a cell’s internal workings. However, cells exist in a multidimensional universe where their gene activity changes with environmental factors and the influence of neighbouring cells.
Traditional sequencing methods use processes that destroy some of these dimensions; they blend tissues into a uniform soup, or extract single cells which might vary from others in the sample. Thankfully, a new approach called spatial transcriptomics (ST) is on the rise: a molecular profiling method that allows scientists to track gene activity in the context in intact tissue.
“ST provides much higher resolution of what’s happening in tissues such as the cancer microenvironment,” said Joe Yeong, a Group Leader at the Institute of Molecular and Cell Biology (IMCB). “That way, we can design better drugs that target the root cause of cancer or address treatment-resistant tumours.”
Analysing ST data requires the use of computational platforms that process expression patterns from groups of neighbouring cells, called clusters. However, current gold standard platforms, such as BayesSpace, can’t handle large volumes of samples or complex tissues with diverse clusters that create 'noisy' datasets.
Yeong’s team came together with scientists from the Duke-NUS Medical School, Singapore; the National Cancer Centre, Singapore; and the East China Normal University, Shanghai, China to take ST computation to new heights. Together, they created SC-MEB (Spatial Clustering using the hidden Markov random field based on Empirical Bayes), a computational tool specifically developed to address the sluggish processing speeds and scalability issues of conventional ST algorithms.
The researchers tested SC-MEB in a series of comprehensive simulations using both human and mouse brain tissues. They found that SC-MEB outperformed its predecessors in terms of clustering accuracy and ran 200 times faster than BayesSpace. SC-MEB was also scalable, handling large datasets with ease without compromising accuracy.
According to Yeong, what sets SC-MEB apart is its ability to classify cell types based on a ‘distance relationship’. “It’s a simple concept,” explained Yeong. “The neighbours of a tumour cell are highly likely to be other tumour cells.” This enables the algorithm to remain agile even while processing big datasets.
The team also performed an additional ST analysis in a tumour sample from a colorectal cancer patient who had COVID-19. They found three COVID-19 gene signatures that were prevalent in immune cells within the colon tissue, but not in other cell types. This further reinforced SC-MEB’s ability to investigate the structural organisation of tissues using transcriptomic data.
Moving forward, the team is expanding the ST data analysis toolbox with more custom-built algorithms. “We put these algorithms together as a standard pipeline and are working with industry partners to provide a one-stop, hassle-free solution for analysing ST data,” said Yeong.
The A*STAR-affiliated researchers contributing to this research are from the Institute of Molecular & Cell Biology (IMCB)