Research activities

This project addresses the general question of inferring interactions in biological systems from massive data and extends recently developed methods in neural coding to the characterization of non-coding RNAs (ncRNAs) and their targets in bacteria from whole transcriptome analysis using massive sequencing. Classical methods to validate functional interactions in molecular biology, by overexpression and deletion, are laborious when going beyond the classic model organisms with well developed genetics. In the age of metagenomics, such organisms represent a small and decreasing fraction of sequenced microorganisms. Alternative procedures, cheaper and more amenable to automation, are therefore of substantial need, as addressed by this project. The project includes validation against now published work to characterize ncRNAs in the bacterium E. faecalis V583, a major cause of hospital-acquired infections. Bacterial growth under multiple, carefully selected, stress conditions provides differential expression for ncRNAs that can be integrated using statistical physical inference methods in order to obtain novel biological information. It will circumvent the problem often encountered in screening for ncRNA functions, i.e. the absence of major phenotypic responses under fixed, standard growth conditions by integrating measurements from experimental assays performed in a catalog of different environmental conditions. 

This project aims at contributing to the fundamental understanding of small systems driven away from thermal equilibrium. Examples of such systems are molecular motors in biology, and micro-manipulation of individual molecules in nanotechnology. Rather than restricting the theoretical analysis of such systems to just an instantaneous system state, a powerful approach developed in recent years considers the entire system evolution and relating a forward process and a backward (time-reversed) process. Fluctuation relations generalize to single small systems (molecules) and to far from equilibrium the classical concepts of fluctuation-dissipation theorems and Onsager reciprocity. The key concept is to externally apply (in theory or in experiment) a time-dependent protocol and study the response behavior of the system by monitoring physically relevant quantities such as applied work, dissipated heat, or the resulting non-equilibrium distributions. This project focuses on the important and common situation where a protocol "control" which optimizes a specific behavior in the system response is of particular interest. The specific goal of this project is to develop a general theoretical framework for optimal protocols and to use this theory for studying the impact of optimal protocols on (thermodynamic) efficiency in small systems, biological and artificial molecular motors, and on work relations and fluctuation theorems. 

The last decade has seen a fruitful interchange between statistical mechanics and problems in information theory, artificial intelligence and computer science. This project has the twin aims to strengthen the Swedish presence in this area and to introduce the program of statistical mechanics of inverse problems in relation to optimization and reconstruction, whence the project title. Important background to the project are recent advances in combinatorial optimisation and large-scale adoption of "message-passing" schemes. Statistical mechanics and physical reasoning are absolutely crucial to the analysis and understanding of these new algorithms, with applications to coding, compressed sensing, inference and many other concrete problems in networks and communications. On the other hand, reconstruction and inference by the maximum entropy criterion, which are very reasonable both from the viewpoint of physics and of information theory, is in a real sense the inverse of a statistical mechanics problem, where model parameters are to be determined from observables (magnetisations, correlation functions, etc.). The project is over four years, and will address the following four goals: (i) to construct local search for max-entropy reconstruction, (ii) to compare recent maximum likelihood schemes to maximum entropy, (iii) to evaluate and extend the dynamic cavity method, and (iv) to develop inference schemes based on fluctuation-dissipation relations out of equilibrium.