Research

We develop computational methods and models to reconstruct phylogenetic trees using maximum likelihood principle. This helps to resolve various evolutionary relationships across the Tree of Life. Highlighted methods include:

• An effective tree search algorithm, which was independently shown to outperform other phylogenetic tools.
• A fast and accurate model selection method (ModelFinder) to find the best common, mixture (MixtureFinder), and partition models (like PartitionFinder).
• An ultrafast bootstrap approach (UFBoot) to assess branch supports.
• Concordance factor, gCF and sCF to quantify conflicts between genomic regions and sites.
• An approximate Bayesian framework (IQ2MC) to infer time trees.

We work on efficient algorithms to infer evolution within very short timescales (e.g. in terms of years) to study emergence of variants of viruses and pathogens, such as the SARS-CoV-2 virus that caused the COVID-19 pandemic, which help to inform public health decisions. Such datasets are characterised by short and similar genomes but very many of them, in the order of millions for SARS-CoV-2, posing a major computational challenge. Our tool CMAPLE takes just a few days to accurately analyse 1,000,000 SARS-CoV-2 genomes, whereas existing methods may take years to complete.

We work on High Performance Computing algorithms to deal with ultra-large genomic datasets. We employ multi-core CPUs and multi-node compute clusters to reduce the runtimes and memory footprints. We want to utilise GPUs to further accelerate computations with lower energy to promote green computing. Our ultimate aim is to fully exploit heterogenous clusters of many CPUs and GPUs.

We are working towards renewing many phylogenetic problems using AI and machine learning. The aim is to speed up, improve, and embed with existing maximum likelihood machineries. In this process, we need to avoid simulation bias and make the AI models generalisable to real data. Our recent work includes IQ-NET to quickly estimate quartet trees using neural networks and ProtFinder to find best-fit protein models using deep neural networks and random forests.

We implement all methods in user-friendly software tools to quickly reach out to the scientific community, enabling faster discoveries. A exemplar outcome is the widely used IQ-TREE phylogenetic software with thousands of users and millions of downloads. IQ-TREE has been continuously developed since 2011 by an international team from Australia, Austria, Canada, Germany, UK, US, and Vietnam. Recently, it has received a lot of code contributions from the community via its GitHub repository. Other software is listed here.

Collaborating with biologists is an important part in our research, facilitating the quest to resolve many questions across the Tree of Life. This is not only to apply our tools to empirical data, but also to identify and overcome potential limitations. The figure illustrates a recent study in Nature, which used advanced models in IQ-TREE to decipher the root of all living organisms characterised by cells with a nucleus, including plants, animals, and humans.