PMC:1679804 / 72990-80231 JSONTXT

{"project":"2_test","denotations":[{"id":"17118189-8697238-1688077","span":{"begin":247,"end":248},"obj":"8697238"}],"text":"Two-step approach for structured pattern search\nSo far we have described the direct method used by SMOTIF to search for the structured motif by positional joins over the symbols. In fact, SMOTIF, can also follow a two-step approach like in Anrep [4] and SMARTFINDER [6]. In the first step, given S MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBamrtHrhAL1wy0L2yHvtyaeHbnfgDOvwBHrxAJfwnaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaWaaeGaeaaakeaaimaacqWFse=uaaa@3845@ and ℳ MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBamrtHrhAL1wy0L2yHvtyaeHbnfgDOvwBHrxAJfwnaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaWaaeGaeaaakeaaimaacqWFZestaaa@3790@, we search S MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBamrtHrhAL1wy0L2yHvtyaeHbnfgDOvwBHrxAJfwnaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaWaaeGaeaaakeaaimaacqWFse=uaaa@3845@ for each simple motif in ℳ MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBamrtHrhAL1wy0L2yHvtyaeHbnfgDOvwBHrxAJfwnaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaWaaeGaeaaakeaaimaacqWFZestaaa@3790@, i.e., M1, M2, ..., Mk. This task can be solved by existing pattern matching algorithms. In the second step, we do positional joins on the pos-lists of the simple motifs. Let P MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBamrtHrhAL1wy0L2yHvtyaeHbnfgDOvwBHrxAJfwnaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaWaaeGaeaaakeaacqqGqbauaaa@3786@(Mi [li, ui] T MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBamrtHrhAL1wy0L2yHvtyaeHbnfgDOvwBHrxAJfwnaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaWaaeGaeaaakeaaimaacqWFtepvaaa@3847@) be an intermediate pos-list, with simple motif Mi as the head, and a suffix structured motif T MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBamrtHrhAL1wy0L2yHvtyaeHbnfgDOvwBHrxAJfwnaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaWaaeGaeaaakeaaimaacqWFtepvaaa@3847@ = Mi+1 ⋯ Mk as tail. Since P MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBamrtHrhAL1wy0L2yHvtyaeHbnfgDOvwBHrxAJfwnaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaWaaeGaeaaakeaacqqGqbauaaa@3786@(Mi) stores only the start positions, we need convert them into end positions to check the gap constraints. There are two cases to consider.\n\nExact matching\nMany algorithms [11-14] exist for exact pattern matching. Like in SMARTFINDER we use a lazy suffix tree [11] to extract the pos-lists for all simple motifs. The matching occurrences are sorted after extracting them from the suffix tree to obtain the pos-list in sorted order. For an intermediate pattern Mi [li, ui]T MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBamrtHrhAL1wy0L2yHvtyaeHbnfgDOvwBHrxAJfwnaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaWaaeGaeaaakeaaimaacqWFtepvaaa@3847@, each start position s ∈ P MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBamrtHrhAL1wy0L2yHvtyaeHbnfgDOvwBHrxAJfwnaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaWaaeGaeaaakeaacqqGqbauaaa@3786@(Mi [li, ui]T MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBamrtHrhAL1wy0L2yHvtyaeHbnfgDOvwBHrxAJfwnaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaWaaeGaeaaakeaaimaacqWFtepvaaa@3847@) is converted into an end position s + |Mi| - 1. Figure 7(a) shows an example of the pos-list join using exact matches for simple motifs. Each column shows the pos-list for a simple motif in the structured motif from Table 4. We first join the pos-lists of TTA and CAT, checking for gap range [1,4]. The start position 8 ∈ P MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBamrtHrhAL1wy0L2yHvtyaeHbnfgDOvwBHrxAJfwnaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaWaaeGaeaaakeaacqqGqbauaaa@3786@(TTA) is converted to end position 8 + 3 - 1 = 10. We find that both positions 12 and 15 lie within the minimum and maximum gap range (indicated by the links), and thus 8 is retained in the resulting pos-list. Likewise 5 is in the final pos-list, since after obtaining its end position 5 + 2 - 1 = 6, we find d = 8 - 6 - 1 = 1 ∈ [0, 1].\nFigure 7 Simple Motif Positional Joins Example. The figure shows an example of positional joins on simple motifs, using the motif from Table 4.\n\nApproximate matching\nSeveral algorithms [8, 12, 15–18] exist for approximate pattern matching. For consistency, we used Sellers' dynamic programming algorithm [26], as implemented in SMARTFINDER, to extract the pos-lists for all simple motifs with approximate matches. This algorithm is not optimal and it can be replaced by more efficient ones [16-18]. Since we allow a specific Levenshtein distance [8] (i.e., insertions, deletions and substitutions) between the occurrences and the motif, the length of the occurrences can be different from the component length |Mi|. Thus we augment the pos-list to explicitly store the end position, in addition to the start position, for each occurrence. Figure 7(b) shows how the pos-list joins work for approximate matches of simple motifs. In the structured motif from Table 4, we consider the exact matches of GC and CAT, and the approximate matches of TTA within Levenshtein distance of 1. Each column in (b) shows the pos-list of a simple motif: the left sub-column is a list of its start positions and the right sub-column is a list of its end positions. We first join the pos-lists of TTA and CAT, checking for gap range [1,4]. We compare the end positions of TTA and the start positions of CAT and find that the pairs (9,12), (10,12), (10,15), and (11,15) all lie within the gap range (indicated by the links), and thus the pairs, (7, 10), (8, 9), (8, 10) and (8, 11) are retained in the resulting pos-list. Likewise (5, 6) is in the final pos-list, since after comparing the end position of GC, 6, with the start position of TTA, 7 and 8, we find d = 7 - 6 - 1 = 0 ∈ [0,1] and d = 8 - 6 - 1 = 1 ∈ [0, 1]."}