Ultra-large alignments using phylogeny-aware profiles

Table 1 Comparison of two UPP variants on representative full-length datasets with respect to alignment SP-error, tree error, and TC scores

Model condition	Method	Alignment SP-error	ΔFN	TC score
10 AA	UPP (Default)	24.2	3.4	11.4
10 AA	UPP (Default, No Decomp)	24.5	5.2	11.0
ROSE AA	UPP (Default)	2.9	1.8	2.6
ROSE AA	UPP (Default, No Decomp)	2.8	1.4	2.5
CRW	UPP (Default)	12.5	7.8	1.4
CRW	UPP (Default, No Decomp)	13.3	16.5	0.9
HomFam (19)	UPP (Default)	23.0	NA	46.6
HomFam (19)	UPP (Default, No Decomp)	25.4	NA	44.5
Indelible 10000M2	UPP (Default)	3.5	0.6	1.2
Indelible 10000M2	UPP (Default, No Decomp)	3.3	0.5	1.4
Indelible 10000M3	UPP (Default)	1.3	0.2	4.6
Indelible 10000M3	UPP (Default, No Decomp)	1.3	0.1	4.8
Indelible 10000M4	UPP (Default)	0.3	<0.0	27.4
Indelible 10000M4	UPP (Default, No Decomp)	0.5	<0.0	30.5
RNASim 10K	UPP (Default)	9.5	0.8	0.5
RNASim 10K	UPP (Default, No Decomp)	11.2	3.0	0.3

All criteria (errors and scores) given as percentages. See text for explanation of names of methods and computational platforms used. The default setting for UPP is denoted UPP (Default); it uses a backbone of size 1000 and uses PASTA to compute the backbone alignment and the ensemble of HMMs technique. In the “No Decomp” versions of these two methods, the ensemble of HMMs is replaced with a single HMM. ML trees are estimated using RAxML (on the 10 AA datasets) or FastTree (all other datasets) except for HomFam, where we do not estimate ML trees as there are no reference trees for the HomFam datasets. NA Not applicable

ISSN: 1474-760X