Efficient pairwise RNA structure prediction and alignment using sequence alignment constraints

BMC Bioinformatics

Table 2 Varying Pin Selection Posterior Cutoff. Effects of varying the posterior probability cutoff at a fixed window length of 20 nucleotides, for all pairs of tRNA sequences in the test set.

Selection Criteria	Num Pairs	Lowest % ID	Full Sankoff		Constrained		Time (s)		Memory (MB)
			BP Sens	Align ID	BP Sens	Align ID	Avg	Longest	Avg	Longest
tRNA subset
no pins	184	9	77 ± 2	83 ± 2	-	-	570	2674	255	560
= 1.0	12	56	88 ± 3	92 ± 3	88 ± 3	92 ± 3	97	197	137	226
≥ 0.99	115	35	84 ± 2	94 ± 1	84 ± 2	94 ± 1	34	218	69	236
≥ 0.98	124	18	84 ± 2	93 ± 1	84 ± 2	93 ± 1	32	240	72	236
≥ 0.95	135	14	83 ± 2	91 ± 1	82 ± 2	91 ± 1	32	171	70	232
≥ 0.90	146	10	81 ± 2	88 ± 2	80 ± 2	87 ± 2	33	145	72	232
≥ 0.85	152	10	80 ± 2	87 ± 2	79 ± 2	86 ± 2	31	241	69	232
≥ 0.80	158	9	80 ± 2	86 ± 2	79 ± 2	85 ± 2	31	139	69	214

For higher posterior thresholds, there are fewer sequence pairs in which at least one pin can be identified; the number of sequence pairs that can be pinned and aligned is indicated in the Num Pairs column, and the percent identity of the most dissimilar pair that can be pinned is indicated in the Lowest ID column. For the subset of pairs that can be pinned at a given posterior threshold, we evaluate basepair sensitivity (BP Sens) and alignment identity (Align ID) for full Sankoff (unconstrained) alignment and the Constrained (pinned) alignment. Comparing these numbers within each row indicates the performance impact of the constraints. The average and maximum resource requirements (Time and Memory) are also shown for each posterior cutoff. The performance gain from pins is primarily determined by the number of pins assigned in a sequence pair. For the subset of sequence pairs that can be pinned at all, the average number of pins is largely a function of the window length and somewhat independent of posterior threshold (here W = 20; avg pins = 2.6). The maximum time and memory requirements are typically for a sequence pair in which only a single pin was discovered.

ISSN: 1471-2105