Input data type for best results
#589
by retract - opened
Using a new dataset of single-cell RNA or pseudobulked RNA data, is it best practice to use data similar to the pretrained corpus for embedding extraction? I tried raw absolute expression and fold change expression, not sure which embeddings are the most accurate outputted by the model
Thank you for your question. Single-cell raw counts data (not pseudobulked, no feature selection) should be used as described here: https://huggingface.co/ctheodoris/Geneformer/blob/main/examples/tokenizing_scRNAseq_data.ipynb
ctheodoris changed discussion status to closed