Approach

Step 1.
Baseline classical ecological network for each system


The NGB project will learn ecological networks from NGS data in 6 distinct ecosystems. For each system we have either a known ecological network or an established set of expectations for the structure of the network.
The systems also reflect the broad combination of drivers of ecosystem change, scales of organisation, and biomes that NGB approach could applied to.
Stacks Image 79
Quantitative interaction network between syrphid pollinators (left) and flowering plants (right).
Step 2.
Sample and sequence eDNA in each study system
and identify Operational Taxonomic Units (OTUs)


Each ecosystem will be sampled using established techniques to give replicate eDNA samples across test situations, which include the natural variation in the presence/absence of particular OTUs and the variation due to drivers of ecosystem change.
The eDNA samples from all systems will then be centrally sequenced at the Genome Transcriptome facility in Bordeaux. Standard OTU databases and bioinformatics pipelines will then be used to assign the sequences in the samples to OTUs.

Stacks Image 87
eDNA is sequenced and then parsed through bioinformatic pipelines to identify known OTUs present in the sample.
Step 3.
Learn network structure in OTU data using Machine-Learning


Statistical and logical inference techniques will be used to reconstruct the system networks, based on the methods previously used for ecological networks, but with appropriate development for NGS data. Tree-based methods and latent Gaussian models will be used to reconstruct networks statistically. Logic-based learning will be done primarily using Inductive Logic Programming (ILP), with Meta-interpretative Learning (MIL) being used to infer generic network interaction rules. The relative performance of the logic-based and statistical learning will be benchmarked for speed and quality of reconstruction of networks from NGS data.
Stacks Image 97
Microbial network of the oak tree (Quercus robur L.) susceptible to the foliar fungal pathogen, Erysiphe alphitoides (Ea). Each node represents a microbial OTU and red and green links indicate hypothesised co-exclusions and co-associations, respectively.
Step 4.
Compare learnt and classical ecological networks


Statistical network comparison analyses will determine the similarity between machine learnt networks, reconstructed from NGS data, and already existing networks or structural expectations from classical ecological sample data, for each system.
Stacks Image 99
Network-core size and composition analyses for comparing two networks.


  Systems