Open Targets Hackathon 2025 Winner

BiomarkerFinder

Discover biomarkers faster. Understand them better.

Open Targets evidence for any disease, with AI explanations clearly labelled.

About BiomarkerFinder

Audience Choice Award Winner

The Achievement

Audience Choice Award

BiomarkerFinder won the Audience Choice Award at the Open Targets Hackathon 2025 – an AI tool built to make biomarker insights explainable for everyone.

Our Mission

Our mission was to make biomarker data more explainable and accessible to clinicians, researchers, and students by combining Open Targets data with AI-powered plain language explanations.

The Team

Nick Lamb PhD, CMPP
Project Lead
Alireza Hekmati
Team Member
Kristian Gurashi PhD
Team Member
Mutanu Malinda
Team Member
Rishika Gupta
Team Member

Learn More

Visit Open Targets Hackathon
Quick Start

How It Works

Three simple steps to discover, understand, and export biomarker insights

BiomarkerFinder

Find the genes most strongly linked to any disease, the evidence behind each link, and the approved drugs that act on them.

Mantle cell lymphoma
Non-small cell lung cancer
Breast cancer
Colorectal cancer
Acute myeloid leukemia
Processing your request...

Search for biomarkers

Enter a cancer type or disease name above to discover relevant biomarkers

Select two or more diseases to visualize shared and unique targets. This can help uncover hidden relationships and potential new therapeutic avenues.

Knowledge Graph will appear here...

Built at a hackathon, hardened afterwards

BiomarkerFinder began as an entry to the Open Targets Hackathon 2025, where it won the Audience Choice award. In October 2026 it was revisited and checked against its own evidence:

  • AI-guessed labels removed. The first version asked a language model to call each gene diagnostic, prognostic or predictive. Checked against the evidence, those labels were inconsistent (four mismatch-repair genes in colorectal cancer received two different labels), so they were replaced with evidence tags read directly from Open Targets.
  • Score relabelled. The 0 to 1 score is Open Targets' association score, not a measure of biomarker utility, and now says so.
  • Drug graph repaired. An Open Targets schema change had silently emptied the drug network. It now uses the current API and shows approved drugs.
  • Backend locked down. Calls are restricted to this site, rate-limited and capped, with API keys held in a secrets manager.

The AI still writes the plain-language explanations, but it is given the Open Targets evidence and told not to classify.

Got Questions?

Frequently Asked Questions

Everything you need to know about BiomarkerFinder

A biomarker is a biological molecule — such as a gene, protein, or genetic mutation — that provides information about a medical condition. It can help detect a disease, predict how it may progress, or indicate how well a treatment is working. In cancer, biomarkers are used for diagnosis, prognosis, and guiding treatment decisions.
We use data from the Open Targets Platform, a public database that connects genes, diseases, and drugs. Our tool sends real-time queries to the Open Targets GraphQL API to get the most updated information.
The score (0 to 1) is Open Targets' overall association score. It combines evidence from genetics, somatic mutations, drugs in clinical development, pathways, literature and animal models into one measure of how strongly a gene is linked to a disease. A higher score means more and stronger evidence. It is not a measure of how useful a gene is as a clinical biomarker.
Each gene is tagged with the types of evidence Open Targets holds for its link to the disease:
  • Drug-response biomarker: the gene is listed in the Cancer Biomarkers database as affecting response to a drug.
  • Clinical drug target: a drug acting on this gene has reached clinical trials or approval for the disease.
  • Somatic mutation: mutations in the gene are found in tumour tissue for this disease.
  • Genetic association: inherited variants in the gene are linked to the disease.
The tags come directly from Open Targets data, not from AI. They show what kind of evidence exists, not whether a gene is used clinically as a diagnostic, prognostic or predictive test.
The gene list, scores and evidence tags come straight from Open Targets. The short explanation on each card and the voice summary are written by an AI model (OpenAI GPT-4o mini), which is given the Open Targets evidence for each gene and told not to overstate clinical use. AI explanations can still contain errors, so follow the "See the evidence on Open Targets" link to check the underlying data.
After clicking "Show Full Analysis," you can toggle to a network graph showing how the disease connects to its top associated genes and to drugs acting on them. Each gene is coloured by its strongest type of Open Targets evidence (drug-response biomarker, clinical drug target, somatic mutation or genetic association).
Yes! You can export the results as CSV or PDF files for easy sharing and record-keeping.
You can type or speak your questions using the microphone button in the search bar. After searching, you'll receive an AI-powered voice summary that explains the biomarker findings in spoken form. You can choose from 4 different AI voices (Rachel, Adam, Bella, Callum), copy the transcript, and optionally expand to see detailed analysis.
Yes! The cross-disease comparison tool lets you analyze multiple diseases at once to find shared biomarkers and drug repurposing opportunities. This helps identify common biological pathways and potential therapeutic targets across different conditions.
No. This tool is for educational and exploratory purposes only. It is not intended to guide clinical practice.