Voi siete qui

AnacletoLab paper among the five best of the year in Medical Informatics26-09-2018

The prestigious recognition for the Bioinformatics laboratory of the Department of Computer Science 'Giovanni Degli Antoni' of the University of Milan comes from the International Medical Informatics Association, for the development of an innovative computational system that allows the identification of gene/disease associations for a large range of human pathologies.

Prediction of Human Phenotype Ontology terms by means of hierarchical ensemble method is the title of the paper written by the AnacletoLab of the Bioinformatics Lab of the Department of Computer Science 'Giovanni Degli Antoni of the University of Milan that was selected by the International Medical Informatics Association (IMIA) as one of the five best "Knowledge Representation and Management" papers of 2017 in the field of Medical Informatics.

The paper, which represents a fundamental step toward the discovery of genes associated with human disorders (to use the words of IMIA's summary), is authored by Marco Notaro, a Computer Science PhD student and Giorgio Valentini, coordinator of AnacletoLab and Associate Professor in the Computer Science Department of the University of Milan, in collaboration with Peter RobinsonJackson Lab for Genomic Medicine (Connecticut, USA) and Max Schubach - Berlin Institute of Health.

The novelty of this research, which was developed in the context of the Monarch Initiative, lies in the integration of biomedical ontologies – in this case the Human Phenotype Ontology (HPO) – with the architecture of machine learning models for the prediction of associations between genes and abnornal phenotypes.

This integration, realized through novel computational methods (hierarchical ensembles of learning machines) enables the discovery of genes associated with pathological human phenotypes, starting from predictions made by state-of-the-art learning algorithms, and improving them in a next step by exploiting the relationships between the biomedical terms of the HPO, which includes more than 13,000 terms (classes of abnormal phenotypes) and more than 156,000 annotations of hereditary diseases.

"This is an important result - says Professor Valentini - because it allows us to systematically improve the predictions of machine learning algorithms for gene/abnormal phenotype associations, when the annotations related to such associations are described with formal ontologies".


*Legend for the Figure

The abnormal phenotypes of the Human Phenotype Ontology are represented as nodes (circles in the figure), connected by edges (arrows). A disease can be characterized as a set of abnornal phenotypes.