Non-negative Autoencoding of 60 Million Single-cell Transcriptomes at Rank 256

This is the gene weights matrix (W) from Non-negative Autoencoding (NNAE) of the human Cell Census, using 60 million single-cell transcriptomes. All healthy human cells from v2025-LTR release were included, and all genes were used, log-normalized. To use this W matrix, ensure that new data is standard log normalized (divide all counts in a cell by the total counts of that cell, multiply by 10000, take log1p) and multiply by W to get cell embeddings. See preprint for details.