The Future of Medical AI - From Physical and Theoretical Limits to Hard Computational Barriers

Future medicine will heavily rely on computational approaches to personalize the treatment of individual patients. In particular artificial intelligence and big data left footprints in the field and yield lots of potential for the future. However, given the many thousand
- sometimes even millions - of molecular parameters (genes, metabolites, proteins, miRNAs, etc.) give thread to overfitting models and to the identification of passengers rather than disease drivers. One of the biggest problems in computational systems medicine, however, is combinatorial explosion, which is inherent in almost all interesting systems/network medicine problems. In the talk, we will exemplify some of those issues (network alignment, sub-network extract, n-cluster editing) and illustrate why efficient solutions to such problems are imperative to elaborate systems medicine, though hard to emerge. We present some initial solutions to these algorithmic issues and relate them to the future development of AI in biomedicine.