Gene tree incongruence is a major predictor of the differences between parsimony and maximum likelihood in phylogenomics

SANTIAGO A. CATALANO; AMBROSIO TORRES; GOLOBOFF, PABLO A.

Berkeley

Encuentro; XXXVIII Meeting of the Willi Hennig Society; 2019

Willi Hennig Society

In previous studies we found a high congruence between the results of Maximum Parsimony (MP) and Maximum Likelihood (ML) for 149 phylogenomic datasets (average of 2.4 SPR-moves; 65% of the resulting trees from MP and ML were identical or differed in one SPR-move), with those differences generally not affecting the main conclusions discussed in the original studies. In addition, we found that most of the incongruence between MP and ML was related to poorly supported nodes. Furthermore, in preliminary analyses we observed an association between branch lengths/missing data and the differences between the results of both methods. In this study we performed a more detailed evaluation of the incongruence between MP and ML, exploring other characteristics that can be related to the MP-ML incongruence, and improving the statistical comparisons. We evaluated whether the amount of incongruence between MP and ML was related to the taxonomic ranks to which the terminal taxa belong and whether the incongruent nodes present specific locations in the trees (basal vs. terminal). Finally, using a multiple linear regression approach we assessed whether there was a relationship between the MP-ML incongruence and the incongruence among gene trees. We found that gene tree incongruence was significantly associated to the MP-ML incongruence, showing that incongruent nodes are in general those that were recovered in a low percentage of the gene trees. We did not observe a significant relationship between the taxonomic ranks and the MP-ML incongruence. The same occurs with the location of the incongruent nodes, where we found no specific trend. The results obtained in the present study indicate that gene tree incongruence is an important explaining factor of the MP-ML incongruence. However, our findings evidenced that explaining the incongruence between parsimony and likelihood in phylogenomic inference requires taking into account several factors in an integrative approach.