Fig. 1

The study design and main results. a In this prospective cohort study, patients suspected of having indeterminate pulmonary nodules were enrolled and their CT information was recorded. Blood samples were collected and stored prior to surgery. TCR sequencing was performed and analyzed. b Venn diagrams were used to identify clones just existing in the malignant group. common motif of enriched CDR3β aaSeqs is displayed. c The top 10,000 TCR clones are shown by Packed circles for the benign and stage I groups (one circle represents one clone; circle size represents clonal fraction). d The relationship between the clone types and the clinical characteristics of the benign and malignant groups is demonstrated. e A boxplot to illustrate the difference between benign and malignant clones for small-type and hyperexpanded-type clone types. f Comparison of TCR repertoire diversity between benign and malignant groups using boxplots. g Comparison of the diagnostic value of TCR repertoire diversity with previous models using ROC curves. h Boxplot showing the difference in Shannon index between benign and malignant groups in the GGN group and the non-GGN group separately. i Random forest and information gain methods were used to select significant features for the TCRnodseek model. j The predicted value calculated by TCRnodseek in the discovery and validation groups are presented as boxplots and ROC curves (k). l the ROCs of the models with and without TCR features. m the relationship between TCRnodseek predicated value and various clinical characteristics. n A Sankey diagram illustrating the misclassification of TCRnodseek and experts in comparison with pathology results (Thoracic_doc, the results determined by Thoracic doctors; CT_doc, the results determined by Radiology doctors; TCRnodseek_youden, the results determined by TCRnodseek with youden cutoff). o A bubble plot illustrating the relationship between CT doctors’ diagnoses, pathology results, and TCRnodseek predicted value (the size of the bubble represents the mean predicted value of TCRnodseek; the color of the bubble represents the number of subjects in the condition). p A waterfall plot showing the TCRnodseek function for nodule sizes <= 20 mm (the X-axis represents the sample id sorted by the predicted value of TCRnodseek; the Y-axis represents the predicted value of TCRnodseek minus the cutoff value). q The diagnostic performance of TCRnodseek in the validation group (the original subjects of (b) to (i) figures are all subjects; the original subjects of (l) to (q) figures are independent validation subjects; GGN ground glass nodule)