This section shows the analysis of ESM-2 embeddings including domain clustering and logistic regression classification results.
This section shows the results of the EmbedDiff latent diffusion model training and synthetic sequence generation.
Figure 3: Logistic Regression Confusion Matrix (ESM-2)
Figure 4: Diffusion Training Loss (ESM-2)
Figure 5: Generated Embeddings t-SNE (ESM-2)
Figure 6: Transformer Decoder Loss (ESM-2)
Figure 7: Real-Real Cosine Similarity (ESM-2)
Figure 8: Generated-Generated Cosine Similarity (ESM-2)
Figure 9: Real-Generated Cosine Similarity (ESM-2)
Figure 10: Identity Histogram (ESM-2)
Figure 11: Entropy vs Identity Scatter (ESM-2)
Figure 12: All Histograms (ESM-2)
Figure 13: t-SNE Domain Overlay (ESM-2)