The added value and even necessity of Artificial Intelligence (AI) becomes clearer every year. AI simulates what people already do well (like chess, driverless cars, recognizing patterns et cetera). In our initiative, The Discovery Network on AI and Cancer, we want it to contribute to health, biology and cancer in particular. That is a lot more complicated.
In healthcare AI is already pretty good in radiology. AI is exceptional in pattern recognition, therefore it enters the world of radiology. But the biology of patients is different and difficult. What can it contribute to this field of expertise? When we are looking for added value in this area we need to penetrate and understand the biology of patients. It is not sufficient to replace this black box by another one, namely the workings of the learned algorithms. Trey outlines this clearly in the 2018 article on Using Deep Learning to model the hierarchical structure and function of a cell.
We are looking for the integration of different datasets. Datasets with information on all the levels from molecular processes in the cell, about tumor cells, tissue, RNA, organoids, drug response and valuable other clinical data to population statistics and lifestyle data. We probably will also need information of citizens that do not have the disease (yet) and have similar lifestyles. Information about resilience. Analysis of these data requires very subtle and sparse patterns to be recognized, and to be biologically or medically relevant these discovered patterns need to be more or less “proved true or extremely likely” rather than just deemed 95% probable.
In this Discovery Network we aim to bring together biologists, mathematicians, experienced machine learning experts, epidemiologists, and perhaps others?