Simulate Phase II population composition before site selection
Simulate Phase II population composition from available real-world evidence to pre-identify high-risk subgroup proportions before site selection and protocol lock.
How it worksVirtual patient modeling for clinical development
See which patient subgroup your candidate fails in
Valinor Discovery builds virtual patients from multimodal cohort data, so a development team can locate the failing subgroup before enrolment opens.
Most trial failures are subgroup failures.
Phase II and III failures in oncology and rare disease routinely trace back to heterogeneous patient populations masking the candidate's true responder profile. The subgroup signal was present in the data. The tooling to surface it was not.
~90%
of Phase III oncology failures involve unexpected population heterogeneity
$1.2B
median sunk cost per Phase III program that fails at readout
3-7 yrs
typical time lost per late-stage attrition event before restart
From cohort data to subgroup failure predictions
01
Ingest
EHR time-series, genomic panels, lab metadata, and phenotypic annotations from CDISC SDTM, HL7 FHIR, and standard trial data formats
02
Synthesize
Probabilistic virtual patient cohort generation from aggregate cohort statistics and literature-curated epidemiological priors
03
Simulate
Trial population simulation across candidate exposures. Identify which subgroup signal collapses, and at what enrichment threshold it recovers
Built on probabilistic patient modeling
Each virtual patient is a statistical sample drawn from a Bayesian network trained on cohort-level conditional distributions. No individual patient record is replicated. The model learns dependencies between biomarker axes, comorbidity patterns, and prior treatment exposure, then samples a population with the right epidemiological shape for your development context.
Bayesian Networks
Conditional probability structures encode the known biological dependencies between biomarker axes, comorbidity prevalence, and treatment response patterns.
Multimodal Integration
EHR time-series, genomics, lab panels, and phenotypic annotations are fused into a unified patient representation before sampling begins.
Privacy by Architecture
No individual patient records are stored or reproduced. The platform trains exclusively on aggregate cohort statistics derived from de-identified source data.
Two critical junctures. One modeling framework.
Rank candidates by predicted performance heterogeneity
Rank candidate targets by predicted performance heterogeneity across modeled patient subgroups before committing to IND-enabling study budgets.
How it worksTechnical capability
What the platform ingests and what it returns
Data Inputs
CDISC SDTM / ADaM structured trial datasets
HL7 FHIR patient records (de-identified, aggregate-only)
VCF / genomic panel outputs (WES, targeted sequencing)
Lab time-series: CBC, metabolic panel, proteomic markers
ICD-10 coded phenotypic annotation files
Simulation Outputs
Synthetic cohort composition reports (n = 1,000 to 10,000 virtual patients)
Subgroup failure probability distributions by biomarker axis
Recommended enrichment criteria for trial protocol design
Target-population alignment score per candidate compound
Built by researchers who ran the trials
The Valinor Discovery team brings backgrounds in computational biology, biostatistics, and clinical development, not software engineering seconded into life sciences.
Josh Pacini
CEO & Co-Founder
Led Phase I and II development operations across oncology and rare disease programs before co-founding Valinor Discovery.
Dr. Priya Mehta
CTO & Co-Founder
Computational biologist specializing in Bayesian modeling of patient populations and probabilistic cohort reconstruction from real-world evidence.
Dr. Marcus Webb
CSO & Co-Founder
Translational biologist focused on patient subgroup analysis, population-level biomarker variability, and Phase I-II trial design.
Join our early-access program
We are working with a small number of translational and clinical development teams on Phase I-II compounds with patient heterogeneity challenges. If that describes your program, we want to hear from you.
Angel-backed, Boston-based team. Honest about stage.