hōʻike - A Joint-Embedding Predictive Architecture for Transcriptome Data Generation with Diffusion Models

Silico Biosciences

Transcriptomic data generation with diffusion in a JEPA-based framework.

Data Preparation

Multiple datasets are used to generate the training data. These include:

• GTEx V10 RNA-seq gene expression dataset: https://storage.googleapis.com/adult-gtex/bulk-gex/v10/rna-seq/GTEx_Analysis_v10_RNASeQCv2.4.2_gene_tpm.gct.gz

• Various TCGA RNA-seq Projects: https://portal.gdc.cancer.gov/