Disease progression
Structures clinical events around progression endpoints and supports research of risk patterns across defined time horizons.
Explainable ML platform for researching cancer progression risk
A research system for assessing progression risk and analyzing therapy response over 3, 6 and 12-month horizons, based on available clinical data.
Request accessDesigned for longitudinal clinical data
The platform builds a patient timeline: diagnosis, therapy, follow-up assessments, response and confirmed progression. This format makes it possible to study disease dynamics, not only isolated medical records.
Research ML platform modules
Therapy response
Studies associations between clinical indicators and the risk of non-response or early progression during the current line of therapy.
Patient stratification
Groups cohorts by shared clinical patterns to support exploratory analysis of early-progression risk.
Imaging context
Imaging-derived features linked to clinical and molecular context. On the roadmap; not part of the first MVP.
Proprietary prognostic ML system
M1
Temporal modeling
Analyzes the sequence of clinical events and estimates progression risk at defined time horizons.
M2
Comparative validation
Compares several prognostic approaches and selects the configuration with stronger robustness, calibration and portability.
M3
Explainability and uncertainty
Shows the factors that influenced the result and reports when the available data are insufficient for a confident assessment.
Data foundation
First models trained on open, curated oncology cohorts
METABRIC
Molecular Taxonomy of Breast Cancer International Consortium — a large curated breast-cancer cohort with clinical and molecular annotation.
TCGA-BRCA
The Cancer Genome Atlas, Breast Invasive Carcinoma — a public multi-omic breast-cancer dataset used for cross-cohort signal validation.
Accessed via cBioPortal cbioportal.org
These open cohorts are Western-derived. They establish the initial signal; validation on regional patient populations is the next step and no regional performance is claimed until then.
Platform
Longitudinal timeline
Raw records are normalized into a per-patient event sequence for research modeling of progression and therapy-response dynamics.
Risk bands under validation
Low, medium and high bands for 3, 6 and 12 months are treated as research outputs and must be validated before pilot use.
SHAP explanations
Research outputs can include contributing factors such as stage, Ki-67, lymph nodes and marker trends, so assumptions can be inspected.
Privacy-first deployment
Designed for de-identified data, local or private-cloud deployment and controlled data access during research pilots.
Uncertainty assessment
The interface is designed to indicate when data are incomplete, for example missing HER2 status or no recent follow-up assessment.
Reproducible pipelines
Traceable assumptions, versioned experiments and logged research outputs support auditability during model development.
ReportCohortData
SYNTHETIC_PATIENT · breast · Stage III · ConceptProgression-risk research report concept
3-MONTHLow
6-MONTHMedium
12-MONTHHigh
Ki-67 high
Nodes (N2)
Marker trend ↑
Subtype (HER2−)
// DEMO INTERFACE. Values and patient are synthetic. The model is under development and validation.
Fantom BioLabs development principles
PrincipleStatus
Explainable research outputSHAP factors and inspectable assumptions for each research report
Built into the designUncertainty is shown explicitlyIncomplete data should be visible instead of hidden behind false precision
Built into the designWorks with available clinical dataNGS and ctDNA may be useful but are not required for the MVP concept
MVP scopeLongitudinal clinical datasetsDesigned for retrospective, de-identified treatment histories
In developmentPrivacy-first research workflowLocal or private-cloud deployment options for controlled pilots
Design principleComparison with a clinical baselineMandatory check before pilot use
Being checkedResearch Use Only positioningNot for diagnosis, therapy changes or replacement of clinical judgment
RequiredWhy we are building Fantom BioLabs
Fantom BioLabs team
“Regional oncology centers accumulate years of clinical data, but those records are rarely used for systematic analysis of treatment dynamics. Our task is to turn de-identified archival histories into testable prognostic models that help researchers and clinicians better understand progression risk and uncertainty in the result.”Fantom BioLabs teamRESEARCH USE ONLY
Request access
We work with research and clinical teams on retrospective pilots, patient stratification and evidence generation for a research MVP.
Request accessResearch Use Only. Fantom BioLabs is a research decision-support tool under development. It does not diagnose, does not change therapy and does not replace the clinician. It must not be used to alter treatment for an individual patient before full clinical validation.