Figure 2.

LAR-Sig concordance in LAR identification. (A) The distributions of kappa statistics observed between LAR-Sig and four clustering approaches (TNBCtype, TNBCtype-4, NMF, and fuzzy clustering), for the seven NAC datasets. Median kappas is provided as the darker and longer line, with mean kappa statistic provided with a shorter and lighter line. Concordance was observed to be lowest among comparisons to the refined TNBCtype-4 and also observed between the other clustering methods and TNBCtype-4, suggesting overfitting occurred in the initial efforts. (B) Alluvial diagram depicting the classification agreement among the original TNBCtype, the refined TNBCtype-4, LAR-Sig and pathological complete response to NAC intervention. We observed a high degree of LAR samples failing to achieve pCR, which was evaluated further.