Training (top): single-cell RNA-seq summed into synthetic bulks (grey) and bulk RNA-seq (green) differ
in gene expression by platform. Ladon is trained on both and maps them to a shared representation. A deconvolution model learns from the representations of synthetic bulks with known cell-type proportions.
Inference (bottom): the composition of a new bulk sample is inferred from its representation.
Step through it with the buttons.