FindingPheno
FindingPheno is creating an integrated computational framework for hologenomic big data, providing the tools to better understand how host-microbiome interactions can affect growth and other outcomes.
Understanding the hologenomic domain is a fiendishly difficult problem, with a complex tangle of interactions at many molecular levels both within and between organisms. FindingPheno aims to solve this problem, developing a unified statistical framework for the intelligent integration of multi-omic data from both host and microbiome to understand biological outcomes.
We apply state-of-the-art mathematical and machine learning approaches taken from evolutionary genomics, collective behaviour analysis, ecosystem dynamics, statistical modelling, and applied agricultural research to give us a truly interdisciplinary perspective towards solving this difficult problem. Our project takes a unique two-pronged approach: combining biology-agnostic machine learning methods with biology-informed hierarchical modelling to increase the power and adaptability of our predictive tools.
The tools created in FindingPheno are expected to significantly improve how we understand and utilise the functions provided by microbiomes in combating human diseases as well as the way we produce sustainable food for future generations.
We apply FindingPheno to real world case studies in sustainable food production.
FindingPheno utilises existing data sets to develop our models. We begin with integrated data, i.e. host and microbiome samples collected together under commercially relevant experimental conditions from chickens, salmon and maize. We then demonstrate the resulting tools against more heterogeneous public data collections for tomatoes and bees, refining our performance against real world applications.
The end goal of FindingPheno is to find the true drivers of phenotype in food production systems to unlock the full potential of microbiome interventions for health and sustainability.
Read more on Cordis about FindingPheno: Unified computational solutions to disentangle biological interactions in multi-omics data.
Go beyond pairwise associations towards causation
We develop methods that go beyond the current paradigm of “pairwise” associations studies by using machine learning, Bayesian statistics and causal models to determine the structure hidden in large multi-omics data sets.
Account for biological heterogeneity
We account for the true dynamic nature of the host-microbiome system by modelling both temporal and spatial changes in the microbiome and their interaction with the host environment.
Include prior knowledge
We develop new hierarchical models to incorporate external information from existing databases and research studies, such as gene or pathway information, previous association studies, and the known evolutionary consequences of genomic and metagenomic changes.
We develop methods that go beyond the current paradigm of “pairwise” associations studies by using machine learning, Bayesian statistics and causal models to determine the structure hidden in large multi-omics data sets.
Go beyond pairwise associations towards causation
We account for the true dynamic nature of the host-microbiome system by modelling both temporal and spatial changes in the microbiome and their interaction with the host environment.
Account for biological heterogeneity
We develop new hierarchical models to incorporate external information from existing databases and research studies, such as gene or pathway information, previous association studies, and the known evolutionary consequences of genomic and metagenomic changes.