The immune system: from pixels to knowledge via graph-based methods
Diego Ulisse Pizzagalli
PhD Candidate and Teaching Assistant
Institute for Research in Biomedicine, Faculty of Biomedical Sciences, USI
Bellinzona campus, Swtizerland
Institute of Computational Science, USI, Lugano campus, Switzerland
DEIB - PT1 Room (building 20)
February 6th, 2020
5.00 pm
Contacts:
Matteo Matteucci
Research Line:
Artificial intelligence and robotics
PhD Candidate and Teaching Assistant
Institute for Research in Biomedicine, Faculty of Biomedical Sciences, USI
Bellinzona campus, Swtizerland
Institute of Computational Science, USI, Lugano campus, Switzerland
DEIB - PT1 Room (building 20)
February 6th, 2020
5.00 pm
Contacts:
Matteo Matteucci
Research Line:
Artificial intelligence and robotics
Sommario
The immune system has a critical role in fighting infections, tumors, and other diseases of primary importance. Hence, it represents a promising target for the development of novel therapeutic strategies. However, the immune system relies on a complex network of cell-to-cell interactions which remains largely unknown. Recently established intravital microscopy methods, allowed to acquire videos of cells while migrating and interacting in living organisms. These videos represent a source of biomedical knowledge to reverse-engineering the immune system and understanding how it works in health and disease. However, extracting knowledge from intravital microscopy data is challenging. This due to the appearance and the dynamic behavior of the cells, which exhibit high plasticity and frequent contacts. Additionally, there is a lack of publicly available intravital imaging datasets. These problematics, limit the application of several computer vision methods for video mining. To this end, we initially made available a dataset of intravital microscopy videos along with their analysis done manually by imaging experts. Then, we represented a microscopy video as a hierarchical graph that models the content, from pixels to biological processes. Finally, applied a variety of semi-supervised machine learning methods to extract knowledge from this graph. Among these methods, we proposed a trainable, graph-based, clustering algorithm that can be highly optimized to analyze images. This combination of microscopy data with computational methods allowed us to describe the complex movement patterns of immune cells, revealing different phases of the early immune response towards influenza vaccination.