September 1, 2022
In rare cases, a cancer cannot be traced back to its tissue of origin using available diagnostic tools, leaving patients and oncologists few options for treatment. A new deep-learning approach from the Garg Lab may help classify these cancers of unknown primary by taking a closer look at developmental gene expression patterns, which are often revived or disrupted in cancer cells. Researchers trained the model on a map of correlations built from two cell atlases, one cataloging gene expression data for different tumor types and the other tracing various developmental trajectories for embryonic cells. The model, described in Cancer Discovery, can identify cancer types with a high degree of sensitivity and accuracy.