Computationally optimizing joint tree inference and Multiple Sequence Alignment method

Inferring a phylogenetic tree is a challenging optimisation problem characterised by an exponentially growing search space as the number of taxa increases. Traditional inference methods rely on simplified models that do not account for insertions and deletions (indels), which compromises the accuracy of evolutionary analysis. The Poisson Indel Process (PIP) is a robust model that incorporates indels, providing a more precise reflection of evolutionary processes. However, the complexity and runtime constraints associated with such models have historically limited their application in phylogenetic tree inference. In this thesis, we introduce a set of optimisation and parallelisation strategies designed specifically to enhance the computational efficiency of a PIP-based inference method. Our approach makes effective use of available computational resources, enabling significant runtime improvements without compromising results. Using a 24-hour computing window and identical resources, we have increased the number of taxa processed in the inference from 100 to 400 on a selected real-world dataset. This scaling up demonstrates the potential of our strategies and brings us a step closer to phylogenetic tree inference under PIP for larger-scale evolutionary analyses.