Medical informatics is the application of advanced information management, processing, analysis and visualisation techniques to healthcare related activities. In this theme we leverage on all of our expertise in complex systems, machine intelligence and knowledge management with the objective of bringing these techniques as close as possible to clinicians, hence maximising their translational potential. Thanks to our expertise in scalable machine learning methods, knowledge extraction techniques, large-scale network generation, analysis and visualisation, ontologies and data integration, we are creating human-friendly and robust data analytic pipelines that are able to embrace the sheer diversity and complexity of clinical and omics data.
A biomarker is defined as a characteristic that may be objectively measured and evaluated as an indicator of normal biologic processes, pathogenic processes, or pharmacologic responses to a therapeutic intervention. Many classic statistical techniques are very effective at identifying individual biomarkers, but given our capacity of generating high-throughput omics data thanks to modern biotechnologies, the main challenge nowadays is the identification of multi-variate panels of biomarkers. We are creating rigorous techniques designed to identify small but highly accurate and relevant biomarker signatures, via a combination of classical statistical techniques and modern machine learning methods coupled with sophisticated information visualisation techniques for easily showing their usefulness.
A very relevant task in medical informatics is the identification of subgroups of patients within a dataset. Many diseases are highly heterogenous, combining multiple genetic/environmental sources, and identifying subgroups of patients is key to understand better these diseases and design appropriate treatments. Classic analysis techniques for patient stratification (e.g. based on unsupervised learning such as clustering algorithms) greatly depend on a predefined measure of patients "similarity", with an infinite number of options and no clear best policy across datasets or patients cohorts. Our advanced knowledge extraction techniques allow us to easily identify the sets of data samples handled by different sub-parts of our supervised machine learning models, leading to new methods of identifying subsets of patients/phenotypes that capture knowledge that escapes classic unsupervised techniques.
Our expertise in machine learning and knowledge extraction techniques allow us to create highly accurate prediction models together with highly interpretable explanations of these predictions, leading the way to the construction of digital clinical assistants for diagnosis and prognosis.
We are using our expertise in semantic data integration and graph mining in order to discover new targets for existing drugs, and hence potentially alleviating of part of the costs of discovering new drugs to the pharmaceutical industry.