Validation of ENLIGHT, an AI predictor of immune checkpoint blockade (ICB) response and resistance, across the treatment span.
Bibliographic record
Abstract
2632 Background: Advanced computational AI algorithms, such as ENLIGHT and DeepPT (Med 2023, Nature Cancer 2024), represent a promising approach to identify predictive biomarkers for cancer therapeutics. Evaluation of ICB response prediction via these algorithms through the full span of pre-treatment, on-treatment, and at progression time points provides a dynamic perspective of response prediction abilities. Methods: A post-hoc analysis of two pan-cancer clinical trials was performed: i) BIO2 is a biobanking protocol of ICB-naïve patients (pts) treated with pembrolizumab (NCT02644369); and ii) The IRIS study (NCT04243720) which enrolled pts who have progressed immediately post ICB. In BIO2, complete, partial response or stable disease for >6 months was classified as responders (R), the rest as non-responders (NR). In IRIS, acquired and primary resistance were defined according to trial protocol. ENLIGHT matching scores were calculated using either transcriptomics from NGS (EMS-NGS), or transcriptomics imputed directly from H&E slides using DeepPT (EMS-DP). The predictive value of EMS was compared to PD-L1 IHC, tumor mutational burden (TMB) and tumor infiltrating lymphocytes (TILs) abundance by IHC, and its trajectory across timepoints was studied. Results: 76 pts from BIO2 (23:53, R:NR), and 37 pts from IRIS (18:19, AR:PR), comprising of 14 tumor types, were analyzed. We first established the value of ENLIGHT as a predictive biomarker using the BIO2 pre-treatment samples. EMS-NGS was a superior predictive biomarker compared with PD-L1 IHC, TMB and TIL abundance, while EMS-DP was comparable (Table). The EMS-NGS scores of responders were significantly higher than non-responders pre-treatment (medians: 0.92 vs. 0.62, p = 1.4e-4). Analyzing the trajectory of the EMS-NGS scores across two additional timepoints reveals that while the scores of non-responding patients remained low (median: 0.62, 0.69, 0.67 for pre-, on–treatment and post-progression, respectively), it is higher among responders (median: 0.92, 0.78 for pre- and on–treatment, respectively). Finally, EMS-NGS was higher among pts with acquired vs primary resistance in IRIS (medians: 0.75 vs 0.59, p = 0.17). Conclusions: In two pan-cancer cohorts, EMS-NGS outperformed conventional biomarkers in predicting ICB response. EMS-DP was comparable to conventional biomarkers and could be calculated directly from H&E slides in a fast, low-cost manner. EMS-NGS values were concordant with response or resistance throughout the ICB treatment course, reflecting the level of the tumor’s vulnerability to ICB inhibition. Further validation of ENLIGHT in larger ICB-treated pts is warranted given these promising results. Clinical trial information: NCT02644369 , NCT04243720 . ROC AUC (p) Sensitivity PPV (cf 30% baseline response rate) F1 Score EMS-NGS 0.74 (0.0003) 61 48 54 EMS-DP 0.64 (0.02) 57 45 50 PD-L1 IHC 0.7 (0.003) 70 40 51 TMB 0.64 (0.03) 39 69 50 TILs 0.6 (0.065) 39 52 44
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.005 | 0.006 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.001 | 0.000 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.001 | 0.000 |
| Open science | 0.000 | 0.001 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.001 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".