Calculator Inputs
Paste known chemistry outcomes and continuous prediction scores. Labels may be line-separated or comma-separated. Scores may be separated by commas, spaces, or line breaks.
Example Chemistry Data Table
This sample represents a chemistry screening assay where score values measure evidence for the positive condition.
| Sample | True Label | Prediction Score | Interpretation |
|---|---|---|---|
| C-01 | 1 | 0.95 | Strong positive evidence |
| C-02 | 1 | 0.91 | Strong positive evidence |
| C-03 | 1 | 0.87 | Positive assay signal |
| C-04 | 1 | 0.78 | Positive assay signal |
| C-05 | 1 | 0.72 | Moderate positive evidence |
| C-06 | 0 | 0.69 | False alarm candidate |
| C-07 | 0 | 0.62 | Borderline negative |
| C-08 | 0 | 0.41 | Likely negative |
| C-09 | 0 | 0.35 | Likely negative |
| C-10 | 0 | 0.18 | Clear negative |
Formula Used
At each threshold, the calculator classifies scores as positive or negative, then computes confusion counts and derived performance metrics.
Threshold Rules
Higher positive mode: predict positive when score ≥ threshold
Lower positive mode: predict positive when score ≤ threshold
Core Rate Equations
Sensitivity (TPR) = TP / (TP + FN)
False Positive Rate (FPR) = FP / (FP + TN)
Specificity = TN / (TN + FP)
Decision Metrics
Precision = TP / (TP + FP)
Accuracy = (TP + TN) / Total Samples
Youden's J = Sensitivity + Specificity - 1
Area Under Curve
AUC = Σ[(FPR(i) - FPR(i-1)) × (TPR(i) + TPR(i-1)) / 2]
The AUC is calculated with the trapezoidal rule across ordered ROC points from the strictest threshold to the loosest threshold.
How to Use This Calculator
- Paste the verified chemistry outcome labels into the first field.
- Paste the model or assay scores into the second field.
- Enter the value that represents the positive chemistry class.
- Select whether higher or lower scores indicate positivity.
- Choose automatic thresholds or define a manual threshold range.
- Press the calculate button to generate ROC metrics and plots.
- Review AUC, best threshold, sensitivity, specificity, and confusion counts.
- Export the full threshold table as CSV or save a PDF report.
FAQs
1. What does the ROC curve show for a chemistry assay?
It shows how sensitivity changes against false positive rate across many thresholds. This helps chemistry teams compare assay discrimination, detect overlap between classes, and choose practical decision cutoffs.
2. What does AUC mean here?
AUC summarizes overall ranking performance. A value near 1 means positives usually score above negatives. A value near 0.5 means the assay behaves close to random separation.
3. How is the best threshold selected?
This file uses the highest Youden's J statistic. That approach balances sensitivity and specificity, making it useful when chemistry screening needs a single threshold without custom error costs.
4. Can I use lower scores as the positive direction?
Yes. Some assays interpret lower cycle values, ratios, or error measures as stronger evidence. Switch the positive direction to lower scores indicate positive class before calculating.
5. Why do labels and scores need matching lengths?
Each score must correspond to one known outcome. If counts differ, confusion totals and ROC points become invalid because the calculator cannot pair predictions with actual chemistry results correctly.
6. Does class imbalance affect ROC analysis?
ROC analysis is relatively stable under imbalance, but threshold choices still depend on the chemistry context. Extremely rare positives may require reviewing precision, prevalence, and false positive burden together.
7. Should I use automatic or manual thresholds?
Automatic thresholds are best for exploratory analysis because they test the observed score values. Manual thresholds are useful when chemistry workflows already enforce concentration cutoffs or reporting bands.
8. What should I export after analysis?
Export the CSV when you need threshold-by-threshold diagnostics in spreadsheets. Export the PDF when you want a quick shareable report containing the summary cards, ROC plot, and confusion matrix.