01
Molecular state
Single-cell expression, regulatory activity, mutations, copy-number changes, and pathway programs.
Computational oncology for carcinomas
A new model of carcinoma biology
Pathelium is building Epithelia AI, a multimodal foundation model of epithelial biology. Learning from single-cell, spatial, genomic, pathology, and organoid data, it is being developed to model carcinoma progression and responses to genetic and therapeutic interventions.
Built for cancer researchers, computational biologists, organoid teams, and translational drug discovery.
The biological problem
Cancer begins with changes we do not yet fully understand
Carcinomas develop as epithelial cells lose the programs that maintain healthy tissue. These changes may affect cell identity, differentiation, polarity, genomic regulation, tissue organization, and interactions with neighboring cells.
Researchers can observe parts of this process through single-cell sequencing, spatial assays, pathology, organoids, and perturbation experiments. The challenge is connecting these measurements into a predictive model of how epithelial disease changes over time.
Pathelium is building that computational layer.
Measured in partsIntegrated over time
Cell stateTissue formGenomeEnvironment
Epithelia AI
A research platform for learning transferable representations across epithelial cells, tissues, disease states, and interventions.
Building a model of epithelial cancer biology
Epithelia AI is being designed as a multimodal foundation model that can learn reusable representations of epithelial biology across data types and experimental systems.
Represent
What state is the system in?
Describe cell identity, tissue organization, disease context, and heterogeneity.
Predict
How might that state change?
Generate testable predictions about future states after an intervention.
Conceptual model
From a biological state to what may happen next
The platform is intended to model heterogeneous responses across a cell population—not only an average expression profile.
Conceptual research framework — not a clinical prediction
1Baseline epithelial state
2Tissue and disease context
3Genetic or therapeutic intervention
ModelEpithelia AI
OutputPredicted distribution of future epithelial states
Possible population outcomes
Cell deathContinued proliferationDifferentiationDrug-tolerant persistenceResistanceMore invasive states
Multimodal foundation
Learning across the biology of a tissue
Epithelia AI is being developed to connect signals measured at different biological scales, from molecular programs inside a cell to tissue architecture and intervention response.
02
Spatial context
Tissue neighborhoods, cell-to-cell interactions, architecture, and the local microenvironment.
03
Pathology and morphology
Histology, organoid imaging, structural organization, dysplasia, and invasion-associated changes.
04
Perturbation biology
Genetic alterations, CRISPR interventions, drugs, combinations, doses, and treatment duration.
05
Experimental systems
Cell lines, patient-derived organoids, tissue samples, and longitudinal measurements.
Research applications
What researchers could understand earlier
These are intended research capabilities under development, not clinically validated tools or treatment recommendations.
Model disease progression
Study transitions from healthy epithelium to dysplasia, invasive carcinoma, and treatment-resistant disease.
Predict perturbation response
Estimate how epithelial cells and organoids may respond to genetic alterations, drugs, and drug combinations.
Reveal resistant states
Investigate small or emerging cellular populations associated with treatment persistence and relapse.
Prioritize experiments
Rank candidate interventions and support decisions about which hypotheses to test next.
Connect models to patients
Study how responses in organoids and other experimental systems may relate to tissue and patient-level outcomes.
Why epithelial biology
Built around the cells where carcinomas begin
Epithelia are not interchangeable surfaces. Each tissue maintains a specialized identity, architecture, and relationship with its environment. A useful model must learn those constraints before it can study how they break down.
Explore the epithelial biologyIdentity and function
Lineage identity, differentiation, and the barrier and secretory programs that define a healthy epithelium.
- Lineage
- Differentiation
- Barrier function
- Secretion
Structure and repair
Polarity, adhesion, tissue architecture, and the coordinated repair programs that maintain a tissue boundary.
- Polarity
- Cell adhesion
- Architecture
- Wound repair
Disease transitions
Inflammation, plasticity, dysplasia, invasion, and partial or complete epithelial-to-mesenchymal transition.
- Inflammation
- Plasticity
- Dysplasia
- EMT
Response ecology
Drug tolerance, resistance, and the changing interactions between epithelial, immune, and stromal cells.
- Persistence
- Resistance
- Immune context
- Stroma
Long-term research direction
Toward a virtual model of epithelial response
Pathelium’s long-term objective is an epithelial virtual cell: a computational model that can represent a baseline epithelial system and simulate how a heterogeneous population may change after a genetic or therapeutic intervention.
This is a research direction rather than a completed capability. Progress depends on carefully linked perturbation data, rigorous evaluation, and experimental validation.
The system should produce
- 1Testable predictions
- 2Population-level response distributions
- 3Mechanistic hypotheses
- 4Uncertainty estimates
- 5Suggested validation experiments
Scientific principles
Built for prediction, interpretation, and validation
Multimodal
Connect molecular, spatial, morphological, perturbational, and experimental data.
Predictive
Generate testable predictions about epithelial state transitions and intervention response.
Interpretable
Relate outputs to genes, pathways, cellular programs, morphology, and possible mechanisms.
Uncertainty-aware
Communicate confidence and identify predictions made outside familiar biological contexts.
Experimentally grounded
Evaluate models with held-out data, controlled perturbations, organoid experiments, and eventually prospective studies.
Who Pathelium is for
Built with cancer researchers in mind
- Cancer biology laboratories
- Organoid research groups
- Computational biology teams
- Translational oncology programs
- Biotechnology companies
- Pharmaceutical discovery teams
Research collaboration
Help build a better model of epithelial cancer
We are interested in working with researchers and organizations studying epithelial biology, patient-derived organoids, perturbation response, carcinoma progression, and treatment resistance.
Research updates
Follow what we are learning
Join researchers, clinicians, and biotechnology leaders following Pathelium’s progress in epithelial foundation models and computational oncology.