Visualization

The pipeline includes structured visualization for both model evaluation and biological interpretation.

Model evaluation

• Confusion matrix
• ROC curves
• Probability distributions

Figure 1: Confusion matrix showing the classification performance of the sequence model on the held-out set.

Figure 2: Receiver Operating Characteristic (ROC) curves for each class, illustrating model discrimination ability.

Biological validation

• Prediction vs EC50 correlation
• Prediction vs Δmax or R²

Interpretability

• Saliency maps
• Integrated gradients plots
• Residue importance summaries

Figure 3: Saliency map highlighting residues whose sequence context most strongly influences NADPH responsiveness predictions.

Figure 4: Per-residue importance scores aggregated across the test set, enabling identification of conserved functional motifs.

Error analysis

• High-confidence misclassifications
• Identification of difficult cases

Interpretation

These plots are designed to answer:

• Is the model accurate?
• Does it align with biology?
• Where does it fail?

Visualization is not decorative; it is diagnostic and interpretive.