Adaptive Biosystems

Living organisms exhibit substantial stochasticity at the microscopic level. Yet they manage to act in a highly organized and well-orchestrated manner whenever they have to, for instance during embryogenesis. We want to understand the molecular architectures and computations that cells employ to communicate, learn and adapt to each other to deal effectively with noise. We develop statistical methods to study living organisms in a data-driven fashion but we are also interested in creating novel biological circuits through rational design. Our research is mainly theoretical but we work closely together with experimental groups from the MPI-CBG and abroad.

Scintific image illustrating promoter activity
Fig.: Reconstruction of promoter activity from time-lapse gene expression data. (a) Fluorescence trajectories are extracted from time-lapse movies through image segmentation. (b) Statistical inference was performed using the DPP algorithm (Zechner et al., 2014) to reconstruct mRNA levels and promoter activity in single cells. The algorithm computes a sampling-based approximation of the Bayesian posterior distribution over inaccessible molecular states (e.g., promoter states) and kinetic parameters (e.g., mRNA half-lives). The shaded areas in the upper panel denote the 5 and 95 percentiles of the posterior distribution and the blue lines indicate maximum a-posterior (MAP) estimates. The lower panel illustrates the posterior probabilities over promoter state (active/inactive).

Future Projects and Goals

Our projects have a highly interdisciplinary character and most of them have both theoretical and experimental components. Current projects include:

  1. Inferential methods. Statistical inference provides the essential link between experimental data and quantitative models. While throughput and accuracy of single-cell technologies are improving rapidly, there is still a lack of efficient statistical algorithms that can properly deal with the high content of recent datasets. We want to develop such algorithms and apply them to reverse-engineer and study molecular networks in living organisms. Our philosophy is to avoid ad hoc approximations wherever possible and to derive efficient analytical schemes by exploiting specific properties of the data and models.
  2. Stochastic modeling. Understanding biochemical networks relies on capable stochastic methods that can account for molecular noise and environmental fluctuations. Markov chain models can capture well the discrete randomness of chemically interacting particles, but their application to realistic scenarios is still hampered by their computational complexity. Most existing methods are therefore restricted to very small networks, or based on radical assumptions about the underlying dynamics. One of our major research interests is to turn stochastic models of molecular networks more large-scale, without compromising their faithfulness in realistic biological scenarios. This involves the development of novel theory as well as simulation algorithms. For instance, we are investigating how one can efficiently simulate small subparts of complex stochastic networks.
  3. Adaptive molecular circuits. To reliably organize across space and time, cells need to exchange, interpret and adapt to signals in the presence of noise – a pattern that is in line with a technical notion of intelligent behavior. In this project, we think of cellular and multicellular systems as networks of agents that use local statistical computations and cell-cell communication to achieve mutual goals. We use the methods from 1. and 2. to investigate the molecular principles enabling “smart” collective behavior in cells, for instance in the context of development. Furthermore, we want to find out, how statistical learning and communication schemes can be forward-engineered using biochemical substrates for the purpose of bottom-up synthetic biology.

Methodological and Technical Expertise

  • Stochastic models of molecular networks
  • Statistical inference & learning
  • Molecular computing & nanotechnology
  • Signal processing & adaptive systems