Table 4.

Tree error rates under low fragmentation for the best placement-based method and two MSA-ML methods. We show FN rates (top) and FP rates (bottom). We show results for the three best performing methods: the best placement-based pipeline (UPP(R)-pplacer), the best MSA-ML method (UPP(R)-RAxML), and a standard MSA-ML method (PASTA-RAxML). Each condition has 75% full-length sequences and 25% fragmentary sequences (which have an average 50% length). The best results for each model condition (within 1%) are shown in boldface. The error rates are averaged over 20 replicates for the simulated datasets.

Method	1000M1	1000M2	1000M3	1000M4	RNASim	RNASim2	16S.M	23S.M
FN Rate:
PASTA-RAXML	0.246	0.181	0.095	0.061	0.186	0.163	0.135	0.137
UPP(R)-RAXML	0.157	0.128	0.094	0.061	0.185	0.163	0.112	0.143
UPP(R)-pplacer	0.215	0.183	0.150	0.112	0.243	0.215	0.192	0.244
FP Rate:
PASTA-RAXML	0.248	0.185	0.100	0.083	0.186	0.163	0.595	0.473
UPP(R)-RAXML	0.160	0.132	0.099	0.083	0.185	0.163	0.584	0.476
UPP(R)-pplacer	0.179	0.146	0.111	0.091	0.208	0.178	0.604	0.517