Artificial intelligence (AI) algorithms such as ENLIGHT and DeepPT represent promising approaches to identify predictive biomarkers for immune checkpoint blockade (ICB). Evaluation of ICB response prediction throughout the span of ICB treatment provides a dynamic perspective of a biomarker’s predictive value. A post-hoc analysis of two pan-cancer trials of patients receiving ICB was performed. Samples were available from the pre-ICB, on-ICB and post-progression timepoints. ENLIGHT matching score (EMS) was calculated using measured transcriptome from sequencing (EMS-NGS), and imputed transcriptome from H&E images using DeepPT (EMS-DP). The predictive value of EMS-NGS and EMS-DP for clinical benefit rate (CBR), progression free survival (PFS) and overall survival (OS) were assessed alongside PD-L1 IHC and tumor mutational burden (TMB). Trajectory of EMS across the ICB treatment timepoints was assessed. 154 H&E slides with matched RNA sequencing from 111 patients, representing 14 tumor types were analyzed. The ROC AUC for predicting CBR was calculated for each biomarker: EMS-DP (0.78), EMS-NGS (0.73), PD-L1 IHC (0.68) and TMB (0.62). Higher EMS-NGS was associated with improved PFS and OS whereas EMS-DP was associated with improved PFS but not OS. Both EMS-NGS and EMS-DP at the on-ICB timepoint remained higher for patients with ongoing response or acquired ICB resistance compared to primary ICB resistance. At the post-progression timepoint, EMS did not differ based on type of resistance. In this pan-cancer study, EMS-NGS and EMS-DP were promising biomarkers of ICB response. EMS values were concordant with response or resistance throughout the ICB treatment course. Further prospective validation of ENLIGHT is warranted.
Park et al. (Mon,) studied this question.