ludvb
commented
1 year ago Hi Le,
Thank you for your interest in our work. H&E images can be converted to the XFuse data format using the xfuse convert image command (you can use the --help flag to see which arguments are accepted). You can then use the generated h5 data files as ordinary experiments that you add to the slides section in your toml config file. I would recommend adding the image slides only for the analysis step, i.e., after you have trained your model on the transcriptomics slides, but they can also be included in training.
I've found that in silico analyses sometimes are sensistive to variations in staining between the transcriptomics and H&E-only slides. If there are considerable batch effects, it may be a good idea to increase the amount of color jitter in the data augmentation by increasing the values of line 33 in xfuse/data/slide/iterator/random_iterator.py.
Hello, Thank you very much for providing this exciting tool! I am interseted in the "In silico spatial transcriptomics", which predict the gene expression from the sole H&E stain images. Now I have 10 slices, and 5 out of them have both H&E stain images and associated gene expression data and others only have H&E stain images. How can I use your tool to predict the gene expression of the rest 5 slices?
Best wishes Le