My research focuses on the development of models and efficient algorithms for mining, integrating and comparing systems-level, large, complex and heterogeneous networked data. I am applying these methods on molecular and clinical data to yield new insight into problems related to human health, and on world trade data for tracking the dynamics of economic systems.
We are flooded with large scale omics data that each represents a partial view on the molecular functioning of the cell. In order to gain novel knowledge, these datasets should be mined collectively rather than in isolation from each others.
We first applied data integration in the context of multiple network alignments, to derive novel protein similarity scores for guiding the node mapping processes [J9].
Recently, we applied data integration on cancer-patient data, yielding new insight into ovarian cancer disease based on modern, systems-level molecular data. These insights include stratification of cancer patients into clinically relevant sub-groups, identification of genes responsible of the onset and the progression of cancer, and proposing drug re-purposing opportunities for personalized patient therapies [C9].
In the cell, molecules (e.g., proteins) do not act in isolation but rather interact with other molecules in order to perform their functions. The same holds in other fields that studies interacting objects, such as in economics, where countries trade with each others.
We developed tools to characterized the wirring patterns of nodes in undirected networks and to relates those wirring patterns to their domain specific annotations. These allow us to discover patterns in the dynamics of the World Trade Network that are indicators of wealth, poverty and economic crises [J6].
We furthered these methods to discover biological functions that are performed through conserved patterns of protein interactions across species [J7].
Recently, we extended these methods to study directed networks [J13].
Comparing biological networks is key to find evolutionary relationships between species, or for transferring annotations across the networks of different species. Since network comparison is computationally intractable, robust and efficient heuristic algorithms are needed.
We first proposed pairwise [J8] and multiple [J9] global network alignment methods.
Since all global network aligners are heuristics, they provide diverging answers to the alignment problem. Recently, we investigated how these aligners could be used in combination in order to better mine biological networks [J15].
Analysing and comparing structures of proteins, in isolation or in complexes, is key for predicting their functions. Such predictions allows bridging the increasing gap between captured proteins and available annotations.
We first proposed methods for comparing the 3D structures of isolated proteins [J1, J2, J5].
Since protein structure comparison approaches are heuristics, they provide diverging answers to the comparison problem. Thus, we proposed a web server to comprehensively comparing the protein structure alignments produced by different methods [J3].
Moreover, since protein does not act in isolation but bind with other molecules (ligands), we developed topological and geometrical descriptors of protein binding interfaces that allow predicting the protein-ligand binding affinity [J4]. This is of first importance when developing new drugs, which should have higher affinities with their target proteins than the natural ligands.