AI for biologists

Very exciting new developments in artificial intelligence, particularly in deep learning, have achieved spectacular performances when applied to biological problems. For instance, the AlphaFold system produces protein 3D models based on their sequences, with an accuracy competitive with experimental structures, and this has greatly expanded proteomes structural coverage. Promising steps have also been taken for determining protein interacting partners and complex 3D structures, predicting the functional outcome of point mutations, and designing new proteins with desired shapes or functions. This summer school will offer an overview of the most recent deep neural network architectures and learning algorithms, and their application to protein-related problems. The participants will acquire theoretical knowledge and practical know-how on data representation, architecture design, and training/testing protocols. We will put a particular emphasis on the specific properties of protein sequences and structures, and on how these properties can be leveraged for improving the learning process. Proteins will be used as “case studies” to illustrate general issues associated with machine learning and the definition of meaningful representations in “digital” biology. The participants will be able to transfer the knowledge acquired during the school to other problems and objects (e.g. variant calling in DNA/RNA sequences from nanopore sequencing data, genotype-to-phenotype mapping…). The school will gather leading scientists coming from different backgrounds, namely biology, mathematics, computer science and physics, and working at the interface between artificial intelligence and biology. It is mainly intended for PhD students and post-doctoral fellows in biology and bioinformatics with some interest in using, understanding and developing machine learning methods.