Find the responders before the trial does.
Many trials fail because the effect is real but diluted — a minority respond, and the average washes them out. From a screening-panel cohort, the engine derives a responder signature with explicit N-of-M logic, so an enriched arm can reach significance at a lower n.
The problem
A real effect can still read as noise. When only a subgroup responds, the trial average dilutes them — and a drug that works for the right patients misses its endpoint. Enrichment changes the arithmetic before the first dose.
Responders washed out
A heterogeneous cohort averages responders and non-responders together — a true effect shrinks toward the mean.
effect · masked
Brute-forcing power
Chasing significance with a bigger n costs patients, sites, time and money — and still may not rescue a diluted signal.
n · expensive
Enrichment changes the math
Stratify to the patients likely to respond and the same effect reaches significance in a smaller, enriched arm.
enriched · powered
How the engine applies
Screening panels in, a responder signature out. The engine reads a candidate cohort's screening panels, derives the marker combination that separates likely responders, and expresses it as an interpretable N-of-M rule you can apply at enrollment.
The logic
A rule a statistician and a reviewer can both read. The responder signature isn't a black box — it's an explicit N-of-M rule over a handful of markers, auditable end to end.
Screening panel
The candidate cohort's pre-enrollment panels — the raw material the signature is learned from.
labs · multi-marker
Responder signature
The marker combination that separates likely responders from non-responders in the cohort.
marker set · directionality
N-of-M rule
A patient qualifies when N of M signature criteria are met — interpretable, not a black-box score.
N of M · thresholded
Power at lower n
Concentrating the effect in an enriched arm lets the same signal reach significance with fewer patients.
effect ↑ · n ↓
Who this is for
Anyone designing a trial that can't afford to miss. Where the asset is promising but the signal is heterogeneous, a responder signature is the difference between a readout and a rescue.
«One asset, one shot»
Moderna
BioNTech
Alnylam
Beam Therapeutics
Relay Therapeutics
Moderna, BioNTech, Alnylam, Beam, Relay
«Stratification strategy»
Pfizer
Roche
Novartis
AstraZeneca
Merck
Pfizer, Roche, Novartis, AstraZeneca, Merck
«An enrichment layer to bring»
IQVIA
ICON
Parexel
Labcorp
Medpace
IQVIA, ICON, Parexel, Labcorp, Medpace
Logos shown are market examples of each segment, to illustrate who the application fits — not Zenlo clients or partners, and no endorsement is implied. This is trial-design support, not a regulatory submission.
What this is — and isn't
Enrichment support, not a companion diagnostic
We derive and document a responder signature from your cohort. We don't run the trial, we don't guarantee its outcome, and we don't issue a regulated diagnostic.
Illustrative application
This page shows how the engine would apply to trial enrichment.
- No named sponsor or asset
- Figures shown are example endpoints
- The real signature comes from your data
The cohort owns the result
The signature is whatever separates responders in your data — or nothing, if nothing does.
- Held-out validation, with significance
- Drivers named; nothing cherry-picked
- Reproducible from the same inputs
Not a companion diagnostic
An N-of-M responder rule is design-stage evidence, not a regulated test or a treatment claim.
- For trial design & stratification
- Regulatory strategy stays the sponsor's
- Zero data retention on model calls
Design your enrichment arm.
Send a screening-panel cohort and we return a responder signature: an interpretable N-of-M rule with held-out performance and an evidence trail your biostatisticians can check.