1 paper
Michael M. Danziger, Bharath Dandala, Viatcheslav Gurev +12
Transcriptomic foundation models pretrained with masked language modeling can achieve low pretraining loss yet produce poor cell representations for downstream tasks. We introduce…