Currently, we exclusively support pretraining the model with human and mouse datasets. In the future, we plan to incorporate the ability to train the model using user-provided datasets from humans or mice. However, due to limitations with the current model architecture, we cannot extend support to datasets from other organisms beyond humans and mice

Presently, our model is tailored for training on single-cell ATAC-seq and single-cell RNA-seq data. We're actively exploring possibilities to expand its capabilities to encompass translation across additional single-cell modalities, such as single-cell proteomics data.

The prediction time varies depending on the file sizes and the available HPC (High-Performance Computing) resources. If there's no job in the queue, predicting scATAC to scRNA for 3,000 cells typically takes around 5 minutes, while the reverse, from scRNA to scATAC, usually takes about 10 minutes.

For human data, we utilize the GRCh38 version. If your data is based on hg19, kindly mark them in the anndata var with the genome attributes, and we will lift them to hg38. This pertains specifically to scATAC-seq data. For mouse data, we use the GRCm38 version.