DeepCellMap is a rare multidisciplinary effort between artificial intelligence, neuroscience, and advanced spatial statistics. The tool can be used on human tissues to show how cells are organised in healthy conditions, as well as in neurodevelopmental disorders and neurodegenerative diseases.
Microglia are key cells in our brain's development, maintenance and immunity. They are also part of the pathological signature of many neurodegenerative and neurodevelopmental disorders, including Alzheimer's and Parkinson's. Mapping the organisation of these cells in a developing human brain is particularly difficult due to the complex patterns in the data - microglia change shape as they colonise the brain, and areas of the brain change quickly from one week of pregnancy to the next. These patterns are also difficult to interpret without tools which can process large amounts of data quickly and efficiently.
The changes in shape of microglia are affected by the information these cells receive from the brain environment, and can reflect the cells’ function. Using DeepCellMap, this study aimed to identify the different distributions microglial phenotypes follow during human brain development, how they interact with each other, and whether some phenotypes are associated with each other against others. In particular, the team looked at cases where brain haemorrhages occurred in foetuses whose mothers contracted COVID during pregnancy. The results show that in these cases, microglial cells are tightly associated with blood vessels, suggesting that microglia may respond to, or influence, changes in blood vessel integrity.
With DeepCellMap, the organisation of cells can be determined in an efficient and thorough manner, showing patterns that the naked eye may not pick up. It can also help scientists study interactions between cell types in tissues. This can lead to better understanding of detailed mechanisms in the brain, and the development of treatments aimed at targeting inflammation in diseases like Alzheimer’s and Parkinson's.
Dr David Menassa says:
DeepCellMap can be applied to any cell type, is highly adaptable and can unravel novel biological patterns in human tissues. Artificial intelligence tools will become the future in neuropathology diagnostic practice and tools like DeepCellMap are just the start of a revolution in expedited and automated histological analyses using deep learning and spatial statistics’.
DeepCellMap has been co-developed by Dr David A Menassa (NDCN), with a team of international experts in machine learning and applied mathematics (Mr Theo Perochon & Professor David Holcman, École Normale Supérieure) and advanced spatial statistics (Professor Thibault Lagache, Institut Pasteur).
The full details of the study ‘Unraveling microglial spatial organization in the developing human brain with DeepCellMap, a deep learning approach coupled with spatial statistics’ have been published in Nature Communications.