CompEpigen
MethylBERT Usage on Plasma cfDNA
Dear @hanyangii , Following your README.md, I used 23 HCC tissue samples and 23 normal tissues to fine-tune the model, and then applied it to 24 HCC plasma samples and 32 healthy plasma controls. The complete processing pipeline I followed is described in the attached file: methylbert preprocess_finetune for both tissue and plasma samples,methylbert finetune using the tissue data,methylbert deconvolute on the plasma samples. However, the model trained on tissue samples failed to accurately predict tumor fractions in the plasma cfDNA data. The predicted values did not distinguish tumor and normal plasma samples effectively. I would be very grateful if you could kindly take a look and let me know whether there might be any issue with how I applied MethylBERT, or if it is possible that this dataset may not be suitable for the tool.
