Work Package 11 (Feudel/Freund, Hillebrand, Dittmar)
"Reconstruction of the microbial/molecular interaction network from ecological model descriptions of measured concentration time series"
Broader background of the proposed research project
Water samples that are collected over a time series from a mesocosm experiment render a sequence of snapshots that reflect dynamical changes of the microbial and molecular community (waxing and waning of concentrations). The concentration dynamics can be resolved down to microbial operational taxonomic units (OTUs) and molecular species level by advanced analytic techniques (next generation sequencing of the 16S rRNA gene for the microbial community, ultrahigh-resolution mass spectrometry for DOM). When applied to the sequence of water samples these analyses yield a wealth of time series reflecting the microbial and molecular concentration dynamics. An explanation of the observed concentration changes from the microbial/molecular interaction and a reconstruction of the association network (Fuhrman 2009, Fuhrman et al. 2015) from the mentioned time series poses a methodological challenge (in view of the multitude of species and the necessarily limited length of the time series). The empirical network reconstruction may be based on a class of generic models (VAR) or on paradigmatic models that are widely used in the context of ecological dynamics. An early ecological model description (Billen et al. 1980) of a single species bacteria/substrate dynamics shall serve as the starting point for many-species generalizations.
Outline for the proposed PhD research project
Time series data of the microbial and the DOM compartment are available for Helgoland Roads and for mesocosm experiments. These data will be provided by WPs 4, 8 and 13 and will be analyzed in collaboration with these associated WPs. The resulting high-dimensional time series will be aggregated to describe bacterial consortia and molecular clusters using recently advanced cluster analysis methods (Röder et al. in preparation). An assessment of resulting clusters will be done in collaboration with WPs 4 and 8 (Teeling et al. 2012). The consensus time series of each cluster will serve as an input to the network reconstruction. Besides measures of undirected pair-correlation we aim also at applying measures detecting directed interactions. The concept of Granger causality is well established for time-discrete linear stochastic processes (VAR) and a model-free counterpart based on transfer entropy exists; unfortunately, the last-mentioned approach works only for sufficiently long time series. Instead we will investigate in how far ecological model descriptions of the interaction between microbial OTUs and DOM clusters can be used to faithfully reconstruct the interaction network. In the setup and development of paradigmatic network models we will collaborate with WP10.