Understanding how the model jointly predicts genes and their expression counts
Core architecture component for generative single-cell modeling
zj(L)
Categorical
Zero-Truncated
Poisson
Purpose: Predicts which gene to select next in the cell sentence
Key Properties:
Example Output:
Purpose: Predicts expression level (count) for the selected gene
Key Properties:
Example Output:
Why Both Heads Are Essential:
Innovation: Unlike discriminative models that only classify, this architecture can generate realistic cell profiles by sampling from both distributions sequentially.
See how the dual decoders work together to generate a cell sentence
Biological Motivation:
Mathematical Form:
Efficiency Benefits:
vs. Separate Models: 2x more efficient than training gene and count predictors separately