New approaches to resolving challenging nodes in the tree of life - NERC GW4 + DTP

Molecular phylogenetics and phylogenomics have revolutionized our view of the history and diversity of life. With the advent of modern sequencing methods, we now have enough genome data to infer an overarching tree of life. But this abundance of data creates major problems for existing phylogenetic methods, creating trade-offs between the complexity of the analysis method and the amount of data that can be analyzed. These have led to debates about the most appropriate analyses, and controversy over key relationships across the tree of life, from the deep relationships among cellular domains to the origins of animals. At the same time, new computational methods such as machine learning have been developed by computer scientists for the analysis of big data. These show enormous promise but have not yet been brought to bear in phylogenomics.

We have recently begun to explore the use of machine learning to infer phylogenetic trees from genome data, and our initial results are very promising. These approaches are based on the idea of simulating large numbers of possible trees and genomic data and then finding those trees that are most compatible with the observed data. We will use established statistical methods, including neural networks, that allow us to do this in a principled way. A key aspect is to calculate appropriate summary statistics from the genomic data that allow us to compare between simulated and observed data sets. A challenge that needs to be addressed is to construct realistic models of evolution that allow us to simulate data sets that are very similar to those observed. You will then apply these methods to three key unanswered questions in the history of life:

• The deepest splits in the animal and plant trees
• The evolutionary origins of plastids in secondarily photosynthetic eukaryotes
• The role of horizontal gene transfer in deep time