Figure 4.

Accuracy of tree refinement. (a) Precision and recall for four iterations of tree refinement. Each iteration contains an ALA by F1ALA and online updating of the tree by TIPars or UShER. (b) Gamma20 log-likelihood and parsimony score for four iterations of tree refinement. Gamma20 log-likelihood was calculated by FastTree2 (reoptimizing the branch lengths with a fixed topology). Tree parsimony score was calculated by UShER. (c) and (d) Linear regression of Gamma20 log-likelihood and parsimony score against precision (c) and recall (d) using trees generated in 0–4 iterations of tree refinement with F1ALA + TIPars and F1ALA + UShER. All regressions, except log-likelihood on precision, are statistically significant (P-value < 0.05). The dashed area shows the 95% confidence interval of the regression. Large R² values (>0.85) are marked in red. The differences of some points are too small to present in the graphs (overlapping), especially those from the third and fourth iterations of tree refinement. (e) Lineage annotation accuracy after tree refinement. “Reference” is the tree built by IQTREE2. Only one iteration of refinement by “F1ALA+X” (TIPars or UShER) is reported. (f) Gamma20 log-likelihood and parsimony score after tree refinement; other details as in (e).