Skip to main content
. 2020 Nov 2;10:18802. doi: 10.1038/s41598-020-75708-z

Figure 2.

Figure 2

WSI score distribution for test, OOD-near and OOD-all sets compared to the actual bulk-measured values (lower values in gray). Scores are computed using the ensemble model for each trait (percent positive tiles by majority vote of the best three). Middle white lines reflect the median slide score. Whiskers extend to 1.5×IQR. (a)–(c) breast test, OOD-near and OOD-all respectively; (d)–(f) lung test, OOD-near and OOD-all (respectively). In the test set, especially worth noting are miR-17-5p in both cohorts, MKI67, FOXA1, MYC, mir-29a-3p and ESR1 in breast as well as KRAS and CD274 (PD-L1) in lung. In the OOD sets most notable are MKI67, FOXA1, MYC, ESR1 and FOXC1 in breast as well as miR-17-5p and miR-21-5p in lung.