Bioinformatics Seminar
Time: 11AM
Venue: Davis Auditorium and Online
15 September 2026
This is a WEHI only event.Same Space, Different Beast: Spatial proteomics with COMET
Ilariya TarasovaWEHI
Tumour cells remodel their microenvironment to exclude immune cells that also drives resistance to immune checkpoint blockade. To identify the genes responsible, the Labat Lab performed a pooled CRISPR screen in mice and profiled the lungs by hi-plex spatial proteomics on the COMET platform (Lunaphore), using a 31-plex panel that includes nine epitope tags combinatorially barcoding (Pro-Code) each knockout (KO) genotype. I will describe the workflow built to turn these images into single-cell data (and mostly developed by Imaging Lab). Segmentation used CellSAM, a foundation model, run as a portable pipeline on the Seqera Platform, yielding over 3,5 million cells across 14 images. Cells were phenotyped in QuPath with a CellTune-based extension that couples tree-based classification with human-in-the-loop review. Because no single model resolves both fine-grained immune subsets and 13 tumour genotypes, we trained two complementary rounds, each with XGBoost and LightGBM; iterative expert correction grew both the size and quality of the training set, reaching 88% accuracy and a macro-averaged F1 of 0.81. These annotations now support comparisons of immune infiltration, cell-type composition and spatial neighbourhoods between KO and control lesions (work in progress).