It is correct to use raw counts as input for pydeseq2. DESeq2 is designed to handle raw count data. The package models the count data using a negative binomial distribution. TPM and RPKM are normalization methods that are not suitable for input into pydeseq2 for differential expression analysis. The statistical methods in pydeseq2 assume a count - based distribution and handle normalization and other necessary adjustments internally. Using pre - normalized values like TPM or RPKM can disrupt the assumptions of the model and lead to inaccurate results.:)