Why we have both pseudobulk and real bulk data?

[Souhatifour]

Both pseudobulk and real bulk data are combined to train the BuDDI model specifically the Variational Autoencoder (VAE).

In my understanding:

By simulating pseudobulks with random cell-type proportions, BuDDI learns to disentangle various sources of variation in the pseudobulk data, which may include cell type-specific gene expression patterns and other latent factors and also to break the dependence between cell-type proportions and other sources of variation.

Real bulk data, which comes from traditional bulk RNA-seq experiments, is integrated with pseudobulk data during the training phase. Real bulk data serves as a bridge between the simulated pseudobulk data and the actual biological data. It brings real-world variability, conditions, and responses into the training process.