A precision-recall (PR) curve plots classifier precision against recall as the decision threshold changes. Precision measures how many predicted positives are true positives; recall measures how many actual positives are found. section 5.1.3
In drug discovery, lowering the threshold for predicting cytochrome P450 inhibition usually increases recall but can reduce precision. The curve exposes this trade-off across possible operating points instead of judging one threshold alone. section 5.1.3
Use the curve to interpret model performance:
- Prefer curves nearer the upper-right corner, representing high recall with high precision. section 5.1.3
- Summarize the precision-recall balance with PR AUC; higher values indicate better performance. section 5.1.3
- Compare against a random baseline: its PR AUC equals the class skew, or the proportion of positive instances. section 5.1.3
CYP inhibition screening
A logistic regression model predicts whether molecules inhibit a cytochrome P450 enzyme. Plotting its PR curve alongside a random baseline shows how well the model identifies known inhibitors while limiting false-positive predictions across thresholds. section 5.1.3




