Abstract
The talk will present our research at the Biomedical Cybernetics Group that I established about three years ago in Dresden. We adopt a transdisciplinary approach integrating information theory, machine learning and network science to investigate the physics of adaptive processes that characterize complex interacting systems at different scales, from molecules to ecosystems, with a particular attention to biology and medicine. Our theoretical effort is to translate advanced mathematical paradigms typically adopted in theoretical physics (such as topology, network and manifold theory) to characterize many-body interactions in complex systems and quantitative biomedicine. We apply the theoretical frameworks we invent in the mission to develop computational tools for systems and network analysis. In particular, in biomedicine we deal with: prediction of wiring in biological networks, combinatorial and multiscale biomarkers design, precision biomedicine, drug repositioning and combinatorial drug therapy. In general, we devise theoretical models of structural organization in complex networks and we leverage this knowledge to create novel and more efficient algorithms and to perform advanced analyses and predictions of patterns in complex systems. This talk will focus on two main theories. Firstly, Minimum Curvilinearity, which is a theory for topological estimation of nonlinear relations in high-dimensional data1 (or in complex networks2) and its relevance for machine learning applications in biomedicine. The new topic on the impact of Minimum Curvilinearity for network embedding in the hyperbolic space will be also treated3. Secondly, we will discuss the Local Community Paradigm (LCP)4,5, which is a theory proposed to model local-topology-dependent link-growth in complex networks and therefore it is useful to devise topological methods for link prediction in monopartite and bipartite5 networks. In particular, we will discuss the impact of this new method for pioneering topological methods for network-based drug-target interaction prediction and repositioning6.
References (* indicates first co-authorship)
1. Cannistraci, C. V., Ravasi, T., Montevecchi, F. M., Ideker, T. & Alessio, M. Nonlinear dimension reduction and clustering by minimum curvilinearity unfold neuropathic pain and tissue embryological classes. Bioinformatics 26, i531–i539 (2010).
2. Cannistraci, C. V., Alanis-Lobato, G. & Ravasi, T. Minimum curvilinearity to enhance topological prediction of protein interactions by network embedding. in Bioinformatics 29, (2013).
3. Thomas, J. M., Muscoloni, A., .., & Cannistraci, C. V. Machine learning meets network science: dimensionality reduction for fast and efficient embedding of networks in the hyperbolic space. (2016). at <http://arxiv.org/abs/1602.06522>
4. Cannistraci, C. V., Alanis-Lobato, G. & Ravasi, T. From link-prediction in brain connectomes and protein interactomes to the local-community-paradigm in complex networks. Sci. Rep. 3, 1–13 (2013).
5. Daminelli, S., Thomas, J. M., Durán, C. & Vittorio Cannistraci, C. Common neighbours and the local-community-paradigm for topological link prediction in bipartite networks. New J. Phys. 17, 113037 (2015).
6. Duran, C., … & Cannistraci, C.V.. Pioneering topological methods for network-based drug-target prediction. Briefings in Bioinformatics (to appear 2017)
