From the team
Technical notes, methodology explanations, and perspectives on Phase II attrition from the Valinor Discovery team.
Latest
Most Phase II failures are not compound failures. They are population failures. We look at how virtual patient cohort simulation changes the calculus before you lock a protocol.
Read articleAggregate cohort statistics from RWE sources rarely arrive in the form you need for trial population simulation. Bayesian network priors let us reconstruct conditional distributions from marginal summaries.
Dr. Priya MehtaRanking therapeutic targets by efficacy signal in an aggregate population obscures the subgroup structure that predicts which programs will fail at Phase II readout. Here is why heterogeneity models belong in target selection.
Dr. Marcus WebbThe signal for subgroup failure lives across data modalities. No single dataset (EHR alone, genomic alone, or imaging alone) has enough conditioning dimensions to surface it. Here is how we bring them together.
Dr. Priya MehtaThe headline number on drug development cost obscures where the losses concentrate. Phase III attrition in oncology and rare disease is almost never a pharmacology failure. It is a population design failure.
Josh PaciniKnowing a subgroup signal exists is not enough. Translational teams need probability distributions over enrichment scenarios, not just a flag. We walk through what a simulation output looks like in practice.
Dr. Marcus WebbIndividual patient records are rarely available at the modeling stage. We describe the statistical approach, copula-based joint distribution sampling with Bayesian priors, for generating valid virtual patient cohorts from aggregate statistics.
Dr. Priya MehtaIn rare disease trials with narrow eligible populations, site selection without subgroup modeling means you will enrol the wrong patients before you find out. Simulation-first changes the decision sequence.
Dr. Marcus WebbA practical guide to data readiness for virtual patient modeling. We cover minimum dataset requirements by modality, acceptable data quality thresholds, and how gaps in coverage affect simulation uncertainty estimates.
Dr. Priya MehtaSynthetic control arms and trial population simulation are often conflated. They address different questions at different stages. Understanding the distinction matters for where you invest modeling resources.
Josh PaciniTarget validation has been a flagship AI use case for years, but most approaches still treat patient populations as homogeneous. We survey the current state and identify the gap that population heterogeneity modeling fills.
Dr. Marcus WebbWe started Valinor Discovery in 2023 to address a gap we kept seeing in translational work: the population model underpinning trial design was built on assumptions, not simulation. This is what we are building.
Josh PaciniThe writing describes the science. The platform is where development teams apply it to their programs.
