Best performance AML Challenge via machine learning technique

A team of researchers of the Johann Bernoulli Institute, University of Groningen, achieved the best performance at the Molecular Classification of Acute Myeloid Leukaemia (AML) Challenge. They used a machine learning technique that delivered a 100% correct prediction of AML cases in the challenge’s test set.

The AML Challenge was organized jointly by the DREAM project ("Dialogue for Reverse Engineering Assessments and Methods") and the FlowCAP project ("Flow Cytometry: Critical Assessment of Population Identification"). The goal is to identify cases of AML based on flow cytometry data. Example data of diagnosed patients were provided to the participating teams who had to hand in predictions with respect to a set of 180 patients whose real diagnosis was unavailable to the teams.

Machine learning technique

For identifying the AML cases the Groningen team applied a machine learning technique. This technique has been developed within the Intelligent Systems group at the Johann Bernoulli Institute as part of the NWO supported research project Adaptive Distance Measures in Relevance Learning Vector Quantization (Admire-LVQ). The team members were:

·         Michael Biehl, Professor in Computer Science

·         Kerstin Bunte, PhD student

·         Petra Schneider, former PhD student at the Johann Bernoulli Institute, now Postdoctoral Researcher at the University of Birmingham, United Kingdom

Further information:
www.cs.rug.nl/~biehl
www.the-dream-project.org
http://flowcap.flowsite.org