Abstract:
Heterogeneity presents many challenges to the modelling process within phylogenetic analyses. However, there are a number of effective techniques that aim to improve the detection and modelling abilities associated with heterogeneity. One relatively unused method focusing on this area is the sliding window (SW) approach. We implemented the sliding window approach as a means to assessing conflicting hypotheses and compare competing models. The method performed phylogenetic analysis on a small window of the sequence, before iteratively sliding along and repeating the analysis within each individual window of sites. The process repeated until all sites had been involved within at least one inference. The exploratory analysis demonstrated a number of strengths and advantages associated with this approach, some specific to the method’s ability in detecting interesting patterns within the data. Likelihood ratio tests indicated the SW approach outperformed the typical complete alignment inference in pinpointing the presence of rate heterogeneity. Further investigation was made into the performance of the SW method, whilst demonstrating an application to real alignment data. The SW approach was implemented to test the conflicting topology hypotheses associated with the conifer chloroplast DNA sequences. The log-likelihood and a distance measuring observed variability (OV-distance) were used to compare competing topologies and to identify potentially influential sites, respectively. These statistics were applied within each window of the SW analysis, allowing comparison across regions of the alignment. The results highlighted the versatility of the SW approach, when assessing specific problems of interest. Whilst the sliding window approach was more computationally intensive than traditional heterogeneity inference tools, the ability to profile and apply multiple inferences across an alignment allowed more insightful and detailed heterogeneity detection within the phylogenetic analysis. Results indicated the approach is undoubtedly a useful tool in detecting general model violations, with the potential to be developed into a powerful inference technique. Keywords: Phylogenetics - Sliding window - Heterogeneity - Rates across sites - Model violations - OV-distance - PartitionFinder.