Fig. 2: Development of the lncRNA-based signature.

A Network plot showing the comparisons of overall survival (OS) and disease-free survival (DFS) between our study and several previously published studies. B Schematic diagram of the workflow for developing a serum-based signature. C Hierarchical clustering and heatmap of 23 differentially expressed lncRNAs from 10 HER2+ mBC patient serum samples resistant (red) to T-DM1 treatment and 10 HER2+ mBC patient serum samples responsive (deep blue) to T-DM1 treatment. Euclidean distance and average linkage clustering were used. Each row represents an individual lncRNA, and each column represents an individual sample. The pseudocolor indicates the lncRNA level from high to low. D Hierarchical clustering correlation plot showing the collinearity of 23 candidate lncRNAs. Correlation matrix heatmap of 23 lncRNAs in the training cohort, where each cell represents the Pearson correlation between the row and column corresponding lncRNAs. The legend shows the color change along with the changes in the correlation coefficient from −1.0 to 1.0. E Four lncRNAs selected by LASSO Cox regression analysis. Two hundred cross-validations for tuning parameter selection in the LASSO model corresponding to Fig. 1D. The two dotted vertical lines mark the optimal values according to the minimum criteria and the 1-s.e. criteria. F LASSO coefficient profiles of the 23 HER2+ mBC-associated serum lncRNAs. The vertical red dotted line indicates the optimal value based on the 1-s criterion, which gives four nonzero coefficients: LINP1, 0.575; EGOT, 0.29; BREA2, −0.64; and LINC01503, −0.42. G qRT-PCR analysis of the expression levels of the 4 lncRNAs in healthy individuals and T-DM1-resistant and T-DM1-sensitive patients. The error bars in the graphs represent the standard deviation (s.d.). *p < 0.05, **p < 0.01, ***p < 0.001. p values were obtained via two-tailed Student’s t tests.