search

digitalcytometry / ecotyper

EcoTyper is a machine learning framework for large-scale identification of cell states and cellular ecosystems from gene expression data.
Other
181 stars 42 forks source link

Could EcoTyper be applied to hepatocellular carcinoma? #1

Closed thestarocean closed 3 years ago

thestarocean commented 3 years ago

The concept of the state of different cells in tumor microenvironment is very intriguing, but we noticed that the EcoTyper did not support hepatocellular carcinoma (HCC), one of the most common carcinomas worldwide. Is there a specific reason HCC is excluded or it will be included in the future? Thanks for your reply.

BALuca commented 3 years ago

We performed several checks when assembling the discovery cohort, including a benchmark of the capability of CIBERSORTx to reliably deconvolve each tumor type. We elected to exclude hepatocellular carcinoma since our epithelial signature didn’t generalize well to that tumor type. For more details, please see the analysis presented in Figure S1L of Luca et al., Cell 2021.

However, it would be straightforward to extend EcoTyper to HCC, whether by using a different signature for the malignant compartment, a dedicated signature matrix defined from HCC scRNA-seq data, or by starting with scRNA-seq data to define states and ecotypes.

Also, we have been able to show in Luca et al., Cell 2021 that the pan-carcinoma cell states and ecotypes generalize to melanoma, a non-epithelial cancer. We consider this an indication that the pan-carcinoma cell states and ecotypes could also generalize to carcinomas beyond the 16 analyzed in the paper. However, additional sanity checks before using the results are highly recommended (e.g. using EcoTyper to evaluate the cell state recovery significance in a scRNA-seq dataset from HCC).