"/>Research area

Biomarker Discovery

Finding signals that change decisions — diagnostic, prognostic and predictive biomarkers, rigorously validated.

0
validated panel
0
validation tiers
0
biomarker types
Overview

What this area is.

A biomarker is only useful if it is reproducible and changes a decision. We discover candidate signatures with disciplined feature selection and machine learning, then guard against over-fitting with cross- and external validation.

Candidates are evaluated by ROC/AUC and survival association, and the best are assembled into compact, multi-omic panels with clear clinical intent.

Tools & technologies

scikit-learnglmnetcaretsurvivalpROCBorutaSHAP
Survival curveOutcome split by biomarker status.
Candidate volcanoDifferential features between groups.
Capabilities

What we do.

Core methods we apply in biomarker discovery.

Feature selection

Stable, interpretable selection from high-dimensional data.

ML classifiers

Models tuned for discrimination and calibration.

Validation

Cross-validation and independent external testing.

Performance metrics

ROC/AUC, sensitivity, specificity and calibration.

Survival association

Linking markers to outcome with Kaplan–Meier and Cox.

Panel design

Compact, multi-omic signatures for real use.

Workflow

From data to insight.

How a biomarker discovery project flows end to end.

01

Cohort

molecular + outcome

02

Select

feature selection

03

Train

ML classifier

04

Validate

CV + external

05

Evaluate

ROC · survival

06

Panel

clinical signature

Visual analytics

Publication-grade figures.

Interactive, live-rendered visualisations used in biomarker discovery.

Survival curveOutcome split by biomarker status.
Candidate volcanoDifferential features between groups.
Sample separationDo candidates separate the classes?
Signature matrixPanel features across patients.
Focus

Where we go deep.

Early detection

Markers that flag disease sooner.

Prognostic signatures

Stratifying risk to guide management.

Predictive biomarkers

Matching patients to therapies that work.

Insights

Questions we answer.

A few of the things people ask about biomarker discovery — and our short answers. Ask CGB-AI for more.

Why is external validation essential?

Models over-fit discovery cohorts; independent data is the only honest test of whether a biomarker generalises.

Single marker or panel?

Panels usually beat single markers on robustness — but must stay compact and interpretable to be clinically usable.

Selected research

Publications in Biomarker Discovery.

Drawn from our full record of 173 papers, filtered to this area.

Browse all publications →

Start a biomarker discovery project.

Tell us the biological question and the data you have — we will map out an approach.

Collaborate with us →