scRNA-seq analysis code for the publication Strohmenger et al., 2025

For each image, the nuclear foci channel was extracted, and background signal was suppressed using a Gaussian blur with a kernel size of 11×11 and a standard deviation (σ) of 1. An enhanced image was then generated by subtracting the blurred background from the original signal and setting all negative values to zero. This process emphasized high-frequency features typical of foci while reducing diffuse background noise.

Foci were identified using a Laplacian of Gaussian (LoG) blob detection algorithm, applied across a scale range of σ = 0.5 to 5.0 with 10 intermediate steps. A detection threshold of 0.1 was used, and only the spatial coordinates of the detected blobs were retained.

Cell boundaries were defined using instance segmentation masks, where each cell was assigned a unique identifier generated by a finetuned Cellpose model. To ensure accurate mapping of foci to individual cells, the masks were slightly expanded using morphological dilation with a 5×5 elliptical kernel, taking care to avoid overlapping cell labels.

Manual annotations were used to assign each segmented cell a class label based on YAP/TAZ/TEAD (?) expression. For each annotated cell, the number of foci detected within its region was then quantified.