Technology - responderlab.com

Technology — Responder Lab

We map the structural geometry of oncology drug response.

Our Responder Atlas uses public datasets to build multiple geometries of pharmacogenomic and response data, treating patient populations as architecture — as shapes rather than sets of variables. The structure of that geometry reveals responder populations and why different patients respond.

15+

Years of development

Geometry

Core methodology

4600+

Drugs modelled

Cancer types

LOO

Validation method

How Responder Atlas Works

STEP 01

Data ingestion

Large public pharmacogenomic datasets — gene expression, drug response, mutation annotations — normalised across cancer types and cell lines.

STEP 02

Geometric mapping

TDA and geometric learning methods map patient populations into structural architectures — revealing subgroup geometry invisible to standard analysis.

STEP 03

Responder identification

Responder subgroups are identified as geometric clusters, defined by molecular signatures, and validated for out-of-sample prediction accuracy.

STEP 04

Clinical translation

Architecture maps, biomarker panels, and trial design implications delivered as a structured report — actionable for enrichment or stratification.

Platform Components

⬡

Geometric Learning Engine

Maps pharmacogenomic data into structural architectures using TDA methods developed over 15 years at the Karolinska Institute. Treats patient populations as shapes — not rows in a spreadsheet.

Core IP

◎

Responder Architecture Maps

Visual and computational representations of drug response geometry across patient populations. Each node is a patient. Cluster separation encodes responder subgroup identity.

Interactive demo available

◈

Biomarker Discovery Module

Identifies the molecular features — mutation enrichment, gene expression signatures, pathway activation — that define and locate responder subgroups within the geometric architecture.

Included in Certificate

◐

Out-of-Sample Prediction

Applies learned response architecture to new patient data to identify likely responders before trial enrollment — validated at 82% accuracy on Phase 3 trial data using leave-one-out cross-validation.

82% validated accuracy

Validation

82%Out-of-sample accuracy

LOOValidation method

The core validation result: 82% accuracy identifying responders out-of-sample from Phase 3 trial data. Not by using biomarkers. Or a mutation panel. The structural geometry of response — learned from the training population, applied to held-out patients.

Leave-one-out cross-validation ensures results are unbiased — the model never sees the patient it is predicting. This is a conservative, gold-standard validation approach appropriate for regulatory and clinical development contexts.

Data Sources

TCGAGDSCCCLEPRISMGEOcBioPortalPharmacoDBCTRPv2

Want to see the platform in action? The free Responder Maps demo shows architecture maps for 18 drugs across all major cancer types — no login, no data required.

See the demo →Order an analysis

Drug trial success prediction

Identifying trial Responders

Predicting trial Responders

Contact Us

Menu