Abstract A002: Comprehensive blood profiling for immunotherapy outcome prediction and longitudinal immune trajectory characterisation
Bibliographic record
Abstract
Abstract Background: Immune checkpoint inhibition (ICI) has revolutionised cancer care, but many patients do not mount anti-tumor activity and most develop autoimmune toxicity. Mechanisms and risk factors underlying ICI response and immune related adverse events (irAEs) are incompletely understood. Thus, patient stratification and targeted irAE treatments are significant unmet clinical needs. Here, we use high-throughput spectral cytometry with machine-learning based analytics to characterise longitudinal immune dynamics under ICI. Patients and materials: 706 cryopreserved PBMC samples from 137 patients consented to the EXACT study (NCT05331066) were utilised. All patients received standard of care adjuvant or advanced ICI for skin or renal cancer. Best overall response was annotated per RECIST 1.1(Responders: CR, PR, SD > 6 months). Patients on adjuvant ICI were designated as no-relapse at > 6 months from ICI initiation. irAEs were graded per CTCAE v5 and grade ≥3 considered severe. Methods: PBMCs were stained with 3 antibody panels comprising 114 markers. Data was acquired on a Sony ID7000 spectral analyser. Systems-level characterisation of 23,906 discrete PBMC subsets per sample was performed using IMU Biosciences’ proprietary machine learning platform. Following data QC, feature selection was refined through titration, variance, and correlation filtering. Predictive PBMC signatures were derived at baseline(BL) and C2 using univariate feature selection with bootstrapping followed by stepwise logistic regression, then validated through 100 iterations of 80:20 cross validation. PBMC types associated with irAE onset(Dev), increasing severity(Inc), and resolution(Res) compared to non-irAE on treatment controls were determined (t-test in a linear mixed effects model). Using these cell types, we then repurposed the Slingshot pseudotime method to derive patient trajectories from BL to Dev, and progression to Inc and/or Res. Results: Benefit(responder/no-relapse) prediction achieved AUCs (mean ± SD) of 0.814±0.11 (BL), and 0.85±0.10 (C2). Severe irAE prediction achieved AUCs of 0.84±0.08 (BL), and 0.82±0.13 (C2). Dev and Inc samples of severe irAEs showed significant enrichment of activated non-classical monocytes, CD4 T, CD8 T, gd T, and NK cells. Dev of non-severe irAEs was indistinguishable from controls. In pseudotime, we found a bifurcating trajectory from BL to severe Dev vs. non-severe Dev. A further bifurcating trajectory distinguished progression from BL to on-treatment, then Dev vs. BL to Dev, then Inc. Res represented a return towards on-treatment controls in both lineages. Discussion: Here, we used high-content PBMC profiling to generate immune signatures predictive of ICI outcome with compelling accuracy. We additionally gain mechanistic insights into irAE development and progression to severity. Our findings highlight the transformative potential of machine learning-powered immune profiling to identify predictors and drivers of benefit and toxicity outcomes under ICI with clear implications for patient stratification and irAE management. Citation Format: Max Emmerich, Duncan McKenzie, Carla Castignani, Jack Bibby, Jennie Yang, Marija Miletic, Laura Marandino, Zayd Tippu, Jonathan Lim, Taja Barber, Stephanie Hepworth, Paul Rouse, Lilian Williams, Kim Edmonds, Justine Korteweg, Serena Vanzan, James Larkin, Tom Hayday, Adam Laing, Samra Turajlic. Comprehensive blood profiling for immunotherapy outcome prediction and longitudinal immune trajectory characterisation [abstract]. In: Proceedings of the AACR Special Conference in Cancer Research: Mechanisms of Cancer Immunity and Cancer-related Autoimmunity; 2025 Sep 24-27; Montreal, QC, Canada. Philadelphia (PA): AACR; Cancer Immunol Res 2025;13(9 Suppl):Abstract nr A002.
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 distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.001 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.000 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.000 | 0.000 |
| Research integrity | 0.000 | 0.001 |
| Insufficient payload (model declined to judge) | 0.000 | 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 teacher head, 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".