Optimization of alignment preprocessing used as input on a residual neural
network for substitution model selection

Vega Méndez, Antón

Optimization of alignment preprocessing used as input on a residual neural network for substitution model selection

License

Citation

Vega Méndez, A. Optimization of alignment preprocessing used as input on a residual neural network for substitution model selection. 2022. handle: http://hdl.handle.net/10230/55050

This citation was generated automatically.

Abstract

When performing phylogenetic tree reconstruction, we often see that the use of an incorrect model of sequence evolution can lead to a wrongly modeled tree, and using established methods to do so can be computationally expensive and prone to errors. ModelRevelator uses deep learning to decide which model of sequence evolution to use for a certain multiple sequence analysis. In this project, I tried to improve the performance of one of ModelRevelator’s neural networks by using a different approach to preprocessing the information contained in multiple sequence alignments: By randomly selecting subsets of the alignments I computed a set of summary statistics summarizing the information contained in each one to improve previous methods.