Key Deliverables & Statistical Frameworks
Time-to-event survival modeling, competing risks, oncology treatment response, diagnostic ROC/PR evaluation, clinical calibration, decision curves, and SHAP explainability.
[DOMAIN 7: CLINICAL BIOINFORMATICS & ML DIAGNOSTICS]
Plot 083 — Kaplan-Meier Overall / Progression-Free Survival Curve
Use Case Application: Estimates time-to-event survival probabilities across biomarker-stratified patient groups with right-censored follow-up.
Required Input Data: Patient-level survival time, event status indicator (1=event, 0=censored), and molecular risk group.
Statistical & Algorithmic Method: Non-parametric Kaplan-Meier product-limit estimator S(t) = prod(1 - d_i/n_i) with Greenwood confidence bands and log-rank test.
[DOMAIN 7: CLINICAL BIOINFORMATICS & ML DIAGNOSTICS]
Plot 084 — Competing Risks Cumulative Incidence Function (CIF)
Use Case Application: Accurately estimates disease-specific relapse or mortality incidence when competing events (e.g., non-relapse death) preclude observation.
Required Input Data: Time-to-first-event data with multi-state event codes (0=censored, 1=primary event, 2=competing event).
Statistical & Algorithmic Method: Aalen-Johansen Cumulative Incidence Function estimator and Fine-Gray subdistribution hazard model.
[DOMAIN 7: CLINICAL BIOINFORMATICS & ML DIAGNOSTICS]
Plot 085 — Multivariable Cox Proportional Hazards Forest Plot
Use Case Application: Quantifies independent prognostic Hazard Ratios (HR) and 95% confidence intervals across molecular and clinical covariates.
Required Input Data: Clinical survival dataset with multiple continuous and categorical prognostic predictors.
Statistical & Algorithmic Method: Semi-parametric Cox proportional hazards regression h(t|X) = h_0(t)*exp(beta^T X) verified via Schoenfeld residuals.
[DOMAIN 7: CLINICAL BIOINFORMATICS & ML DIAGNOSTICS]
Plot 086 — Oncology Patient Treatment Swimmer Plot
Use Case Application: Visualizes individual patient treatment duration, response onset, durable remission, and progression milestones in clinical trials.
Required Input Data: Longitudinal patient timelines containing treatment start/end dates, RECIST response timestamps, and ongoing status.
Statistical & Algorithmic Method: Individual subject-level horizontal duration bars sorted by total time on study with overlaid clinical event markers.
[DOMAIN 7: CLINICAL BIOINFORMATICS & ML DIAGNOSTICS]
Plot 087 — RECIST Tumor Burden Waterfall Plot
Use Case Application: Displays best percentage change in target lesion diameter across every patient in an oncology trial cohort.
Required Input Data: Baseline and nadir sum-of-diameters tumor measurements per patient alongside biomarker status.
Statistical & Algorithmic Method: Best percentage change = ((Nadir - Baseline) / Baseline) * 100 ranked monotonically against RECIST v1.1 thresholds (-30% PR, +20% PD).
[DOMAIN 7: CLINICAL BIOINFORMATICS & ML DIAGNOSTICS]
Plot 088 — Longitudinal Tumor Kinetics Spider Plot
Use Case Application: Tracks individual patient tumor growth or shrinkage trajectories over serial imaging assessments to spot pseudoprogression.
Required Input Data: Serial percentage change in tumor burden across scheduled follow-up weeks for each patient.
Statistical & Algorithmic Method: Piecewise linear longitudinal trajectories anchored at 0% baseline across post-treatment evaluation weeks.
[DOMAIN 7: CLINICAL BIOINFORMATICS & ML DIAGNOSTICS]
Plot 089 — Receiver Operating Characteristic (ROC) Curve
Use Case Application: Evaluates diagnostic biomarker or machine-learning classifier discrimination across all classification probability thresholds.
Required Input Data: True binary clinical outcomes (0/1) and continuous model predicted probabilities or biomarker scores.
Statistical & Algorithmic Method: True Positive Rate (Sensitivity) vs. False Positive Rate (1 - Specificity); AUROC computed via trapezoidal rule with DeLong 95% CI.
[DOMAIN 7: CLINICAL BIOINFORMATICS & ML DIAGNOSTICS]
Plot 090 — Precision-Recall (PR) Curve
Use Case Application: Assesses classifier performance on imbalanced clinical datasets (e.g., rare variants, low-prevalence disease screening).
Required Input Data: Binary ground-truth labels and continuous classifier risk scores.
Statistical & Algorithmic Method: Precision TP/(TP+FP) vs. Recall TP/(TP+FN) across thresholds; Area Under PR Curve (AUPRC) compared against baseline prevalence.
[DOMAIN 7: CLINICAL BIOINFORMATICS & ML DIAGNOSTICS]
Plot 091 — Clinical Risk Model Calibration Curve
Use Case Application: Verifies whether predicted probabilities (e.g., 30% relapse risk) match observed event frequencies in patient cohorts.
Required Input Data: Predicted event probabilities and observed binary outcomes from an external validation cohort.
Statistical & Algorithmic Method: Decile-binned or LOESS-smoothed observed vs. predicted event rates; Brier score, calibration intercept, and calibration slope.
[DOMAIN 7: CLINICAL BIOINFORMATICS & ML DIAGNOSTICS]
Plot 092 — Decision Curve Analysis (DCA) Net Benefit Plot
Use Case Application: Determines whether deploying a diagnostic biomarker model improves clinical decision-making over 'treat-all' or 'treat-none' policies.
Required Input Data: Binary clinical outcomes and predicted probabilities from competing diagnostic models.
Statistical & Algorithmic Method: Net Benefit = (TP/N) - (FP/N) * (p_t / (1 - p_t)) evaluated across clinically relevant threshold probabilities p_t.
[DOMAIN 7: CLINICAL BIOINFORMATICS & ML DIAGNOSTICS]
Plot 093 — SHAP Explainable AI Beeswarm Summary Plot
Use Case Application: Deconstructs black-box machine learning predictions into patient-level feature contributions and directional biomarker effects.
Required Input Data: Trained ML model (XGBoost / Random Forest / Neural Net) and patient feature matrix.
Statistical & Algorithmic Method: SHapley Additive exPlanations (TreeSHAP / KernelSHAP) grounded in cooperative game-theoretic Shapley values.
[DOMAIN 7: CLINICAL BIOINFORMATICS & ML DIAGNOSTICS]
Plot 094 — Multi-Class Diagnostic Confusion Matrix
Use Case Application: Pinpoints specific subtype misclassification patterns and class-wise sensitivity/specificity in diagnostic classifiers.
Required Input Data: Ground-truth class labels and discrete predicted class assignments across K diagnostic categories.
Statistical & Algorithmic Method: Row-normalized contingency table with Cohen's kappa, macro-F1, and balanced accuracy metrics.
[DOMAIN 7: CLINICAL BIOINFORMATICS & ML DIAGNOSTICS]
Plot 095 — PCA Biplot (Scores & Feature Loadings)
Use Case Application: Reveals major axes of variance, batch structure, and the specific genes/metabolites driving sample separation.
Required Input Data: Centered and scaled multi-sample feature matrix (samples x features).
Statistical & Algorithmic Method: Singular Value Decomposition (SVD) X = U * Sigma * V^T projecting sample scores and top eigenvector loading arrows.
[DOMAIN 7: CLINICAL BIOINFORMATICS & ML DIAGNOSTICS]
Plot 096 — Bland-Altman Assay Method Agreement Plot
Use Case Application: Quantifies systematic bias and 95% limits of agreement when validating a new clinical assay against a gold standard.
Required Input Data: Paired quantitative measurements (M1, M2) from two diagnostic assays across identical patient samples.
Statistical & Algorithmic Method: Difference (M1 - M2) vs. Mean ((M1 + M2)/2); mean bias d_bar and 95% Limits of Agreement d_bar +/- 1.96 * SD_diff.
[DOMAIN 7: CLINICAL BIOINFORMATICS & ML DIAGNOSTICS]
Plot 097 — Ternary Compositional Simplex Projection
Use Case Application: Visualizes three-component proportions (e.g., tumor/immune/stroma fractions or wild-type/heterozygous/homozygous states).
Required Input Data: Three-part compositional vectors summing to 1 (or 100%) per sample.
Statistical & Algorithmic Method: Aitchison simplex barycentric coordinate transformation mapping (p1, p2, p3) onto an equilateral triangle.