Virtual Molecular Profiling from H&E
Unlock molecular insights from archival H&E slides without expensive re-sequencing
The Challenge
Archival Samples Without Molecular Data
Hospitals have thousands of archival DLBCL H&E slides from years of patient care, but no RNA-seq or genomic data. Re-sequencing all of them would require substantial investment.
Expensive Companion Diagnostics
Sequencing-based companion diagnostics (RNA-seq, NGS panels) are expensive on a per-patient basis. H&E-based diagnostics would be substantially cheaper, enabling broader patient access.
Trial Screening Bottleneck
Pre-screening patients for biomarker-selected trials requires molecular testing. H&E is already performed—using it for pre-filtering would substantially reduce screening costs.
Lost Tumor Microenvironment Context
Bulk RNA-seq misses spatial context (T-cell exclusion, fibrotic stroma). H&E images capture TME architecture, but extracting quantitative features requires AI.
"We have thousands of DLBCL archival slides from years of patient care, but sequencing them all to validate our biomarker would be prohibitively expensive. We need a more cost-effective way to profile these samples."
The Helomnix Solution
Helomnix uses proprietary attention-based deep learning and pathology foundation models to infer gene expression, mutations, and survival-associated features directly from standard H&E slides. Our models are trained on paired H&E-RNAseq data from DLBCL cohorts, learning morphological patterns that correlate with molecular features.
This enables "virtual molecular profiling" of archival cohorts: screen thousands of H&E slides at a fraction of the cost of RNA-seq, substantially reducing retrospective validation costs. Classify ABC vs. GCB subtype, characterize TME immune infiltration, and identify survival-associated features from H&E alone.
For companion diagnostics, H&E-based tests are substantially more cost-effective than sequencing and use slides already collected during standard care, streamlining FDA approval and market access.
Unique Differentiator
Unlike generic digital pathology platforms, we specialize in hematological malignancies with proprietary H&E-RNAseq paired training data from DLBCL cohorts. Our tumor microenvironment profiling characterizes features associated with CAR-T response by quantifying T-cell exclusion zones.
How It Works
H&E Image Processing
Whole slide images (WSI) are quality-controlled, normalized for stain variations, and tiled into patches. Pathology foundation models extract deep features from each patch.
Attention-Based Classification
Attention-based deep learning aggregates patch features to classify slide-level outcomes. Attention maps highlight influential regions (e.g., tumor-stroma interface).
Molecular Feature Inference
Deep learning models infer genome-scale gene expression, COO classification (ABC/GCB), clinically relevant mutations, and survival-associated features from H&E.
TME Profiling & Validation
Quantify tumor microenvironment: immune infiltration, T-cell exclusion, fibrosis. Validate results on held-out cohorts and provide interpretable attention heatmaps.
Real-World Application
A pharma partner had a CAR-T trial with variable outcomes. They wanted to characterize molecular features from archival trial samples to support a rescue trial design.
Before
Traditional approach: Re-sequence all archival FFPE samples with RNA-seq. Substantial per-sample cost and extended timeline for sequencing and analysis.
After
Helomnix H&E-based profiling: Scanned archival H&E slides, characterized TME features (T-cell infiltration, fibrosis) and ABC/GCB subtype at a fraction of sequencing cost. Substantially accelerated timeline.
Outcome
Identified T-cell excluded phenotype as a candidate resistance mechanism. Designed rescue trial enriching for T-cell infiltrated tumors. H&E-based pre-screening now supports cohort characterization at substantially reduced cost and time.
Value to Your Organization
Retrospective Cohort Savings
Screen thousands of archival H&E slides at a fraction of the cost of RNA-seq. Validate biomarkers without expensive re-sequencing.
Time to Insight
Virtual profiling of archival cohorts takes weeks vs. extended timelines for retrospective sequencing and analysis.
CDx Cost Reduction
H&E-based companion diagnostics are substantially more cost-effective than sequencing-based approaches, enabling broader patient access and reimbursement.
Our Methodology
Data Inputs
- Whole slide images (WSI) in standard formats (.svs, .ndpi, .tiff)
- Clinical metadata (diagnosis, stage, treatment)
- Outcomes data (response, survival, relapse) if available
- Quality control: Tissue area, stain quality, artifact detection
AI/ML Techniques
- Automated quality control
- Stain normalization
- Pathology foundation models (pre-trained on millions of images)
- Attention-based deep learning for slide-level classification
- Cross-validation on independent cohorts
- Explainability: Attention heatmaps showing influential regions
Deliverables
- Inferred gene expression profiles (genome-scale)
- COO classification (ABC vs. GCB) with confidence scores
- Clinically relevant mutation classifications
- Survival risk scores (high/intermediate/low)
- TME profiling: Immune infiltration, T-cell exclusion, fibrosis
- Attention heatmaps for interpretability
- Validation report with ROC curves and performance metrics
Discuss a Translational Application
We welcome discussions about how this approach can support your translational research.