DLBCL CAR-T Response Characterization
Characterize CAR-T response profiles by profiling tumor microenvironment before treatment
The Challenge
CAR-T Failure Rates Are High
CAR-T therapy achieves variable complete response rates in r/r DLBCL, but a substantial proportion of patients do not benefit. The high cost of manufacturing and treatment makes non-response a significant burden for patients and healthcare systems.
T-Cell Exclusion Is Associated with Failure
Patients with T-cell excluded tumors (fibrotic stroma, immune deserts) are associated with unfavorable CAR-T biology. CAR-T cells face barriers to tumor infiltration in these microenvironments. This is visible on H&E but not quantified pre-treatment.
No Pre-Treatment Biomarkers
Current practice: Treat all eligible r/r DLBCL patients with CAR-T, discover unfavorable outcomes post-treatment. Need informative biomarkers to characterize patient biological profiles BEFORE manufacturing CAR-T.
Wasted Manufacturing Costs
CAR-T manufacturing is costly per patient. Treating patients without favorable biological profiles wastes manufacturing capacity and delays access for patients with potentially favorable profiles.
"We treated a cohort of r/r DLBCL patients with CAR-T — a substantial fraction failed. Looking back at H&E slides, patients who did not benefit had obvious T-cell exclusion patterns. We should have characterized TME profiles before manufacturing."
The Helomnix Solution
Helomnix combines H&E image analysis with multi-omics profiling to characterize CAR-T response profiles BEFORE treatment. Here, cohort characterization focuses on describing tumor microenvironment states that influence CAR-T biology, rather than predicting individual patient outcomes. Our proprietary digital pathology models quantify tumor microenvironment features from pre-treatment H&E slides: T-cell infiltration, fibrotic stroma, immune exclusion zones, tumor-stroma architecture.
Multi-omics integration (RNA-seq, scRNA-seq) adds molecular context: immunosuppressive gene signatures, T-cell dysfunction markers, ABC vs. GCB subtype. Combined H&E + omics features enable comprehensive characterization of CAR-T response profiles.
This supports cohort enrichment: characterize which patients present favorable TME profiles (T-cell infiltrated tumors) versus unfavorable molecular context (biologically constrained for CAR-T activity), providing molecular context for translational and clinical research workflows.
Unique Differentiator
We integrate spatial H&E features (T-cell geography) with molecular features (gene expression) — neither alone is sufficient. Our paired H&E-RNAseq DLBCL training data captures this multimodal relationship.
How It Works
Pre-Treatment Data Collection
Obtain pre-treatment H&E slides (lymph node biopsy) and bulk RNA-seq or scRNA-seq from same sample. H&E captures spatial TME, omics captures molecular state.
H&E TME Profiling
Digital pathology AI quantifies: T-cell infiltration density, T-cell exclusion zones, fibrotic stroma area, tumor-immune interface length. Generates TME feature vector.
Multi-Omics Integration
RNA-seq profiling adds: immunosuppressive signatures, T-cell exhaustion markers, ABC/GCB classification. Integrates with H&E features.
CAR-T Response Characterization
Machine learning model trained on a curated DLBCL CAR-T cohort identifies molecular features associated with TME biology. Report: TME profile classification (favorable / intermediate / unfavorable molecular context) with key features driving characterization.
Real-World Application
A CAR-T treatment center sought to characterize patient profiles with favorable TME biology before initiating CAR-T manufacturing for r/r DLBCL.
Before
Standard approach: Treat all eligible r/r DLBCL patients (no biomarker selection). High non-response rate with significant wasted resources on patients who do not benefit.
After
Helomnix analyzed pre-treatment H&E and RNA-seq data from a retrospective cohort. Identified T-cell exclusion patterns and immunosuppressive gene signatures as features associated with unfavorable TME biology, enabling structured cohort characterization.
Outcome
Molecular context from TME characterization supported translational interpretation and informed study design considerations. The enriched cohort showed improved alignment between CAR-T biology and patient TME profiles.
Value to Your Organization
CAR-T Cohort Characterization
Improved characterization of patients with favorable CAR-T biology supports cohort enrichment and translational research workflows.
Manufacturing Efficiency
Avoid manufacturing CAR-T for patients with unfavorable TME profiles, supporting meaningful reduction in wasted manufacturing costs.
Faster Patient Access
Better allocation of manufacturing capacity enables faster access for patients with favorable molecular profiles.
Our Methodology
Data Inputs
- Pre-treatment H&E whole slide images (lymph node biopsy)
- Bulk RNA-seq from same biopsy sample
- ScRNA-seq (optional, for immune cell profiling)
- Clinical metadata (stage, prior therapies, CAR-T product)
- CAR-T outcome (CR, PR, NR, relapse, survival)
AI/ML Techniques
- Automated H&E quality control
- Digital pathology foundation models for feature extraction
- Deep learning models for TME profiling from whole slide images
- T-cell infiltration quantification from H&E
- RNA-seq gene signature analysis (immunosuppressive and T-cell exhaustion programs)
- ABC/GCB classification (COO subtype)
- Multimodal fusion (H&E + omics) with ensemble learning
- Feature attribution analysis for interpretability
Deliverables
- CAR-T response characterization (structured risk assessment)
- TME profile classification (favorable / intermediate / unfavorable molecular context)
- TME profile: T-cell infiltration density, exclusion zones, fibrosis
- Key informative features with importance scores
- H&E attention heatmap highlighting influential regions
- Gene signatures contributing to characterization
- TME profile context to support translational interpretation for patients with unfavorable profiles
- Summary report with molecular and TME characterization for translational review
Discuss a Translational Application
We welcome discussions about how this approach can support your translational research.