NOTEBOOK 06 ← Previous | Next →

Clinical Response Integration

TCR Repertoire Analysis — Notebook 06: Integrated feature matrix, correlation analysis, and composite scoring

Joshua LuthyR + edgeR + tidyverseSynthetic Data2025

Contents

Analysis Overview
Integrated Feature Matrix
Summary Heatmap
Composite Score
Conclusions

01 Analysis Overview

This notebook integrates findings across the full pipeline — diversity metrics, clonal expansion, AUC kinetics, and edgeR differential abundance — into a unified patient-level feature matrix. We explore correlations with clinical response (CR, PR, PD) and build a composite score.

02 Integrated Feature Matrix

All per-patient metrics are aggregated into a single feature matrix for clinical correlation analysis, combining Product diversity, fold changes, AUC kinetics, and top clone dominance.

Figure 1. Spearman correlation matrix across integrated features. Green = positive correlation, Red = negative. Strong positive correlations between AUC, persistence, and clonality confirm that these metrics capture related aspects of clonal selection.

03 Patient × Feature Heatmap

Figure 2. Z-scored patient × feature heatmap. Green = favorable (higher expansion/persistence), Red = unfavorable. CR patients show consistently positive z-scores across all features.

04 Composite Score

A composite repertoire score combining z-scored clonality, AUC, persistence, top clone fraction, and inverted Shannon entropy separates patients by clinical response.

Figure 3. Composite repertoire score per patient. CR patients score highest (most focused Product, strongest expansion, best persistence) and PD patients score lowest.

05 Conclusions

Integrated Findings

1. Product repertoire focusing predicts response — CR patients have the lowest Shannon H, highest clonality, and most dominant top clones.

2. Post-infusion expansion kinetics separate responders from non-responders.

3. The composite score integrating multiple metrics separates clinical response groups better than any single measure.

4. These patterns are consistent across all analytical approaches (descriptive, edgeR, longitudinal, integrated).