<img alt="" src="https://secure.lote1otto.com/219869.png" style="display:none;">
Skip to content

Predictive Biomarker Signature Characterization



The term biomarker signature describes the behavior of a set of biomarkers that define a signature to maximize the prediction performance. We examine the behavior of specific biomarkers as a set that consistently fluctuate together to maximize the accuracy on predicting the disease-related outcome.
How we apply a biomarker signature depends on the prediction problem. A prognostic biomarker signature is used to predict the disease progression, a risk biomarker signature is used to identify sets of subjects that are likely to develop a disease, and a predictive biomarker signature is used to determine the patients that are likely to respond to a particular treatment. Predictive biomarker signatures are used often in oncology to stratify patients with a specific cancer into sub-populations and develop targeted therapies for the diseased population subtypes defined by the biomarker signature.

In this blog, we share an example project that our data science team has worked on supporting this work.  The case study forms part of a new ebook 'Innovative Data Science and Real-World Analytics Approaches in Practice' and we are also delighted to provide the link for download as part of the article. 

Development Challenge
Our client was developing a new drug for complex neurodegenerative disease in pre-clinical development. The drug may be only effective for a particular subgroup of patients. Our client needed to generate a hypothesis on the molecular pathway and the targeted drug activity and identify a biomarker signature defining potential response to the new drug.

Cytel Solution with Advanced Analytical Expertise
Cytel data scientists were engaged to narrow down and define the relationship of a pool of
35 molecular variables :
Step 1: Create visualization tools using Principal Component Analysis ( PCA) to narrow down the candidate biomarkers
Step 2: Refine biomarkers further using unsupervised learning clustering algorithms. The goal was to minimize cluster complexity and heterogeneity.
Step 3: Validate initial findings using a new dataset and semi-supervised clustering.
The analysis produced a biomarker signature that was provided to the client for in-vivo validation.
Using these techniques, the client was able to improve their chances of targeting the most promising subgroup for their therapy.


Are you interested in learning more about applications of data science and real-world analytics techniques? Click the button below to download our new ebook that shares practical examples of how advanced analytics can help to transform drug development.

New call-to-action


Further reading

Applying Biomarker Driven Strategies

Trends in Data Science: Podcast with Ursula Garczarek


contact iconSubscribe back to top