Fig 1.
Overview of the different scoring algorithms.
Panel A shows the traditional cell-based H-score, panel B shows the average threshold method (ATM) score, and panel C shows the pix H-score.
Fig 2.
P-cadherin, PD-L1 and 5T4 IHC datasets.
Panels A, C and E show representative images at 20x magnification with varying levels of P-cadherin, PD-L1 and 5T4 expression, respectively, in tumor resections. Panels B, D and F show the plot of the pathologist H-score versus mRNA transcript level for P-cadherin (n = 30 cases), PD-L1 (n = 24 cases) and 5T4 (n = 21 cases), respectively. The panels also show Spearman’s correlation coefficient along with the p-value and 95% confidence interval. When compared to the P-cadherin pathologist H-score, all P-cadherin DIA endpoints (see S1 Table) yielded positive correlations (Fig 3A–3E). The correlation with the ATM score (Fig 3C) and pix H-score (Fig 3D and 3E) were higher than the correlations with the DIA based H-scores (Fig 3A and 3B). More specifically, the Spearman’s correlation coefficient for HALO and QuPath DIA H-scores were 0.5 (p = 0.005) and 0.39 (p = 0.03), respectively, whereas the Spearman’s correlation coefficient for the ATM score, the VIS pix H-score and the HALO pix H-score were 0.78 (p<0.001), 0.77 (p<0.0001) and 0.88 (p<0.0001), respectively. When compared to the P-cadherin transcript, all DIA endpoints similarly yielded positive correlations (Fig 3F–3J), with the pix H-score exhibiting the highest Spearman’s correlation coefficient (Fig 3I and 3J; ρ = 0.83 and ρ = 0.81, respectively, for VIS and HALO pix H-score; p < 0.0001) followed by the ATM score (Fig 3H; ρ = 0.62, p < 0.0001) and the DIA H-scores (Fig 3F and 3G; ρ = 0.5, p = 0.005 for HALO and ρ = 0.45, p = 0.01 for QuPath).
Fig 3.
Performance of DIA endpoints obtained from P-cadherin IHC images.
Panels A through E show the plots of different DIA endpoints versus pathologist H-score for a cohort of 30 head and neck cancer resections. Panels F through J show the plots of different DIA endpoints versus P-cadherin mRNA transcript for the same 30 cases. Each panel also shows the Spearman’s correlation coefficient between the two quantities plotted in that panel along with the p-value and the 95% confidence interval.
Table 1.
Table lists the results of William’s t-test to test for significant difference in the Spearman correlation coefficients between P-cadherin transcript or pathologist H-score and different DIA endpoints.
Fig 4.
Performance of DIA endpoints obtained from PD-L1 IHC images.
Panels A-E show plots of the different DIA endpoints as a function of the pathologist H-score, while panels F-J show the same as a function of PD-L1 mRNA transcript for a cohort of 24 lung cancer resections. All panels show the Spearman’s correlation coefficient between the two quantities plotted in that panel along with the p-value and the 95% confidence interval.
Table 2.
Table lists the results of William’s t-test to test for significant difference in the Spearman correlation coefficients between PD-L1 mRNA transcript or pathologist H-score and different DIA endpoints.
Fig 5.
Performance of DIA endpoints obtained from 5T4 IHC images.
Panels A-E show plots of the different DIA endpoints as a function of the pathologist H-score, while panels F-J show the same as a function of 5T4 mRNA transcript for a cohort of 21 lung cancer resections. All panels show the Spearman’s correlation coefficient between the two quantities plotted in that panel along with the p-value and the 95% confidence interval.
Table 3.
Table lists the results of William’s t-test to test for significant difference in the Spearman correlation coefficients between 5T4 mRNA transcript or pathologist H-score and different DIA endpoints.
Fig 6.
Empirical approach to assess robustness of pix H-score to spatial sampling.
Panel A shows the breakup of the tumor resection into non overlapping circular regions. Panels B, C and D show the results of the bootstrap analysis for PD-L1, P-cadherin and 5T4, respectively, where the average Spearman’s correlation coefficient between the regional pix H-score estimate from N circular regions and pathologist H-score is plotted as a function of the number of circular regions, where N varies from 1 to 50. The red dashed line shows the Spearman’s correlation coefficient between whole-slide Pix H-score and pathologist H-score for that biomarker. Error bars indicate ± SEM.