Oscar Gaggiotti (School of Biology, University of St Andrews)

Website: https://research-portal.st-andrews.ac.uk/en/persons/oscar-eduardo-gaggiotti

Titre : Interpretable Deep Learning for Population Genomics

Big data and artificial intelligence are buzzwords that pervade most scientific fields and society in general but the extent to which AI methods are being used to analyse big data varies greatly across scientific domains. Although many life-science disciplines have quickly embraced DL approaches, evolutionary biology has lagged behind because of its very strong preference for model-based approaches derived from the rich body of population genetics theory. This reticence is further enhanced by the reputation of DL as black-box models that cannot provide mechanistic explanations for the extremely accurate predictions they provide. However, this last roadblock is being removed by recent advances in “interpretable machine learning” algorithms. In this talk I will present two examples of deep learning applications that implement algorithms aimed at identifying the input variables that contribute the most to predictive power and, in doing so, allow us to make inferences about the mechanisms underlying observed patterns of genetic diversity.