| name | explain-bio-dl-model |
| domain | compbio |
| description | Translates complex deep learning architectures (VQ-VAE, Transformers, Neural ODEs) into standard biological manuscript text, focusing on the biological rationale for the architecture. |
Deep Learning Architecture Explainer
Translate the model architecture details provided in $ARGUMENTS into a manuscript-ready text block suitable for a journal like Nature Computational Science or Bioinformatics.
Follow this specific mapping framework:
- The Biological Problem: Start by defining the biological constraint the model overcomes (e.g., noisy single-cell dropout, missing spatial z-axis resolution, high-dimensional perturbation spaces).
- The Architecture Rationale: Explain why the specific architecture was chosen for this biological data.
- Example: "A Vector Quantized-Variational Autoencoder (VQ-VAE) was implemented to force the highly variable gene expression profiles into a discrete latent space, effectively isolating distinct cellular states from continuous technical noise."
- Data Flow: Describe the journey of the data from the input biological matrix, through the embedding layers, to the final biological prediction (e.g., predicting spatial perturbation responses).
- Validation: End by stating how the model's embeddings are biologically validated (e.g., comparing generated gene embeddings against established Hallmark pathways).
