nonzero median
Could you please explain the term "nonzero median"?
Transcript counts are always positive by definition. My understanding is that each gene in Genecorpus-30M has been detected at least once. Given this, I do not undestand how the median could possible be ever zero. Is it not true that each gene's expression, even after both normalization steps, must be positive?
Thank you for your question. Your question is in reference to the normalization factor for each gene that we use to construct the rank value encodings for input into the Geneformer model. We calculate the median expression value across cells where the expression is non-zero. As explained in the methods section of the manuscript, we opted to use the non-zero median value of expression rather than include zeros in the distribution so as not to weight the value by tissue representation within Genecorpus-30M, assuming that a representative range of transcript values would be observed within the cells in which each gene was detected. This normalization factor for each gene is calculated once from the pretraining corpus and is used for all future datasets presented to the Geneformer model. The provided tokenizer code in the model repository includes this normalization procedure and should be used for tokenizing new datasets presented to Geneformer to ensure consistency of the normalization factor used for each gene.
Manuscript: rdcu.be/ddrx0
Geneformer model repository: https://huggingface.co/ctheodoris/Geneformer