MoTeXII: structured MoTif eXtraction from largescale datasets
 Solon P Pissis^{1}Email author
DOI: 10.1186/1471210515235
© Pissis; licensee BioMed Central Ltd. 2014
Received: 27 November 2013
Accepted: 4 June 2014
Published: 8 July 2014
Abstract
Background
Identifying repeated factors that occur in a string of letters or common factors that occur in a set of strings represents an important task in computer science and biology. Such patterns are called motifs, and the process of identifying them is called motif extraction. In biology, motif extraction constitutes a fundamental step in understanding regulation of gene expression. Stateoftheart tools for motif extraction have their own constraints. Most of these tools are only designed for single motif extraction; structured motifs additionally allow for distance intervals between their single motif components. Moreover, motif extraction from largescale datasets—for instance, largescale ChIPSeq datasets—cannot be performed by current tools. Other constraints include high time and/or space complexity for identifying long motifs with higher error thresholds.
Results
In this article, we introduce MoTeXII, a wordbased highperformance computing tool for structured MoTif eXtraction from largescale datasets. Similar to its predecessor for single motif extraction, it uses stateoftheart algorithms for solving the fixedlength approximate string matching problem. It produces similar and partially identical results to stateoftheart tools for structured motif extraction with respect to accuracy as quantified by statistical significance measures. Moreover, we show that it matches or outperforms these tools in terms of runtime efficiency by merging single motif occurrences efficiently. MoTeXII comes in three flavors: a standard CPU version; an OpenMPbased version; and an MPIbased version. For instance, the MPIbased version of MoTeXII requires only a couple of hours to process all human genes for structured motif extraction on 1056 processors, while current sequential tools require more than a week for this task. Finally, we show that MoTeXII is successful in extracting known composite transcription factor binding sites from real datasets.
Conclusions
Use of MoTeXII in biological frameworks may enable deriving reliable and important information since real fulllength datasets can now be processed with almost any set of input parameters for both single and structured motif extraction in a reasonable amount of time. The opensource code of MoTeXII is freely available at http://www.inf.kcl.ac.uk/research/projects/motex/.
Keywords
Motif extraction Structured motif Transcription factor binding sitesBackground
Identifying repeated factors that occur in a string of letters or common factors that occur in a set of strings represents an important task in computer science and biology. Such patterns are called motifs, and the process of identifying them is called motif extraction. Motif extraction has numerous direct applications in areas that require some form of text mining, that is, the process of deriving reliable information from text [1]. Here we focus on its application to molecular biology.
In biological applications, motifs correspond to functional and/or conserved DNA, RNA, or protein sequences. Alternatively, they may correspond to (recently, in evolutionary terms) duplicated genomic regions, such as transposable elements or even whole genes. It is mandatory to allow for a certain number of errors between different occurrences of the same motif since both single nucleotide polymorphisms as well as errors introduced by wetlab sequencing platforms might have occurred. Hence, molecules that encode the same or related functions do not necessarily have exactly identical sequences.
A single DNA motif is defined as a sequence of nucleic acids that has a specific biological function. The pattern can be fairly short, 5 to 20 basepairs (bp) long, and is known to occur in different genes [2], or several times within the same gene [3]. The DNA motif extraction problem is the task of detecting overrepresented motifs as well as conserved motifs in a set of orthologous DNA sequences. Such conserved motifs may, for instance, be potential candidates for transcription factor binding sites for a regulatory protein [4].
In addition to this simple form of DNA motifs, structured motifs are another special type of DNA motifs. A structured DNA motif consists of two (or even more) smaller conserved sites separated by a spacer (gap). The spacer occurs in the middle of the motif because the transcription factors bind as a dimer. This means that the transcription factor is formed by two subunits having two separate contact points with the DNA sequence. These contact points are separated by a nonconserved spacer of mostly fixed or slightly variable length. Such conserved structured motifs may, for instance, be potential candidates for transcription factor binding sites for a composite regulatory protein [5].
In accordance with the pioneering work of Sagot et al.[6, 7], we formally define the single and structured motif extraction problems as follows.
A single motif is a string of letters (word) on an alphabet Σ. Given an integer error threshold e, a motif on Σ is said to eoccur in a string s on Σ, if the motif and a factor (substring) of s differ by a (Hamming) distance of e. The single motif extraction problem takes as input a set s_{1},…,s_{ N } of strings on Σ, where N≥2, the quorum 1≤q≤N, the maximal allowed distance e (error threshold), and the length k for the motifs. It consists in determining all motifs of length k, such that each motif eoccurs in at least q input strings. Such motifs are called valid.
Each element m_{ i } of a structured motif is called a box and its length is denoted by k_{ i }.
 1.
${m}_{1}^{\prime},\dots ,{m}_{\beta}^{\prime}$ are in s and
 2.
the distance between the end position of ${m}_{i}^{\prime}$ and the start position of ${m}_{i+1}^{\prime}$ in s is in $\left[{d}_{{min}_{i}},{d}_{{max}_{i}}\right]$, for all 1≤i<β.
The structured motif extraction problem takes as input a set s_{1},…,s_{ N } of strings on Σ, where N≥2, the quorum 1≤q≤N, β lengths (k_{ i })_{1≤i≤β}, β error thresholds (e_{ i })_{1≤i≤β}, and β−1 intervals ${({d}_{{min}_{i}},{d}_{{max}_{i}})}_{1\le i<\beta}$ of distance. Given these parameters, the problem consists in determining all structured motifs that have an (e_{ i })_{1≤i≤β}occurrence in at least q input strings. Such structured motifs are called valid.
Related work
 1.
Wordbased methods that mostly rely on exhaustive enumeration, that is, counting and comparing oligonucleotide sequence (kmer) frequencies;
 2.
Probabilistic sequence models, where the model parameters are estimated using maximumlikelihood or Bayesian inference methods.
Here we focus on wordbased methods, since probabilistic sequence models often cannot converge to the global optimum. A plethora of wordbased tools only for single motif extraction, such as YMF [9], Weeder [2], FLAME [10], and MoTeX[11] have already been released. In the search for more complex motifs, fewer methods have been released that extract DNA sites composed by two boxes, such as DyadAnalysis [4] and MITRA [5]. To the best of our knowledge, there exist only two wordbased tools that can address the problem for multiple boxes with distance intervals: RISOTTO [12] (the successor of RISO [7, 13]) and EXMOTIF [14].
Let us first describe the approach used in RISOTTO for single motif extraction. This approach was first introduced by Sagot in [6]. RISOTTO initially indexes the set of N strings using a truncated suffix tree[15]. The suffix tree is then modified to store a boolean array of size N at each node of the suffix tree. This array indicates the strings in the input dataset that contain the factor labeling the path from the root to the corresponding tree node. RISOTTO subsequently searches for eoccurrences of motifs along different paths of the suffix tree. For every valid motif, one has to walk along at most N×n different paths in the suffix tree, where n is the average string length. For every string of length k induced by a path in the tree, there exist at most Σ^{ e }k^{ e } valid motifs, where Σ is the size of the alphabet Σ, and e is the error threshold. Hence, the overall time complexity of this approach is $\mathcal{O}\left(\Sigma {}^{e}{k}^{e}{N}^{2}n\right)$, where the additional factor N is required to access the boolean arrays.
For structured motif extraction, RISOTTO makes uses of an additional data structure, the boxlink. This data structure is constructed to store the information needed to jump from box to box. Informally, a boxlink is a tuple of tree nodes, corresponding to these jumps in the suffix tree. For clarity of description, let us assume that each box has the same length k and a fixedlength gap from the next box. The extraction of structured motifs starts by extracting single motifs of length k, one at a time. The suffix tree is temporarily and partially modified so as to extract the subsequent single motifs. When no errors are allowed, there exist at most Σ^{ β k } ways of spelling all structured motifs. In this case, the total number of visits made to nodes between the root and level k of the suffix tree is bounded by $\mathcal{O}\left(\Sigma {}^{\mathrm{\beta k}}\right)$. However, when up to e errors are allowed in each box, a node at level k may be visited $\mathcal{O}\left(\Sigma {}^{\mathrm{\beta e}}{k}^{\mathrm{\beta e}}\right)$ times more; the total number of visits made to nodes between the root and level k of the suffix tree is $\mathcal{O}\left(N\Sigma {}^{\beta (e+k)}{k}^{\mathrm{\beta e}}\right)$, where the additional factor N is required to access the boolean arrays. A number of operations is also needed to update and restore the suffix tree. In overall, the time complexity of RISOTTO for structured motif extraction is bounded by $\mathcal{O}\left(N\Sigma {}^{\beta (e+k)}{k}^{\mathrm{\beta e}}\right)$.
EXMOTIF uses an inverted index of symbol positions, and it enumerates all structured motifs by positional joins over this index. The distance intervals constraints are also considered at the same time as the joins. Let us first describe the approach used in EXMOTIF for single motif extraction. There exist potentially Σ^{ k } single motifs, and, therefore, in the worst case, $\mathcal{O}\left(\Sigma {}^{k}\right)$ single motifs may be extracted. For a single motif of length k, EXMOTIF uses O(logk) positional joins to obtain the total number of input strings that contain at least one occurrence of the single motif, and each such join takes $\mathcal{O}\left(\mathit{\text{nN}}\right)$ time. Thus, extracting the single motifs takes time O(n N log(k)Σ^{ k }) in the worst case. For Σ^{ k } single motifs, there exist Σ^{ β k } potential structured motifs. When no errors are allowed, extracting the structured motifs requires time O(β n NΣ^{ β k }). However, when up to e errors are allowed in each box, extracting the structured motifs requires time O(β n NΣ^{ β k }+β^{2}k^{ e }Σ^{ e }). Hence, in overall, the time complexity of EXMOTIF is bounded by O(β n NΣ^{ β k }+n N log(k)Σ^{ k }).
Our contribution
All aforementioned algorithms for single and/or structured motif extraction exhibit all or a part of the following disadvantages:

Their time complexity depends on or grows exponentially with the motif length k. Hence, they can only be used for finding very short motifs [16]. For instance, YMF allows only up to k:=8 and Weeder up to k:=12.

Their time complexity depends on the size Σ of the alphabet. Hence, they are not suitable for detecting motifs drawn from large alphabets (e.g., amino acids, where Σ=20).

Their time complexity grows exponentially with the error threshold e. Thus, they are not suitable for detecting long motifs with higher error thresholds, say k:=13 and e:=4.
There are two additional disadvantages:

Existing tools are only designed for identifying motifs under the Hamming distance model (mismatches) but not under the edit distance model (indels). Indels in biological sequences may occur because of insertions or deletions of genomic segments at various genomic locations or due to sequencing errors.

Existing tools are not designed or implemented for highperformance computing (HPC). For instance, Weeder and RISOTTO, which are currently two of the most widely used tools for motif extraction, require more than two months to process all human genes for single motif extraction, with k:=12 and e:=4, making this kind of analyses intractable [11]. A parallel algorithm for the extraction of structured motifs exists [17], but the implementation is not publicly maintained. Moreover, in [16], the authors mention that they plan to improve their algorithm’s ability to process largescale ChIPSeq datasets.
To alleviate these shortcomings, we have introduced MoTeX, a wordbased HPC tool for single MoTif eXtraction [11]. A valid single motif is called strictly valid if it occurs exactly (with no errors), at least once, in any of the input strings. By making this stricter assumption for motif validity, we reduced the problem of single motif extraction in solving the fixedlength approximate string matching problem [18] for all N^{2} pairs of the N input strings. We demonstrated that this approach can alleviate all the aforementioned shortcomings of stateoftheart tools for motif extraction; and produce very promising results both in terms of accuracy under statistical measures of significance as well as efficiency. A part of these wellknown issues for single motif extraction were discussed and addressed in [19] and [20]. Notice that the reduction proposed here makes the time and space complexity of MoTeX not directly comparable to the ones of RISOTTO and EXMOTIF which solve a harder algorithmic problem.
In this article, since also most of the aforementioned tools are only designed for single motif extraction, we introduce MoTeXII, the successor of MoTeX, for the more involved case of structured motif extraction from largescale datasets. To detect the structured motifs, one may apply single motif extraction to detect each box separately. However, this solution breaks down when some boxes are insignificant. Thus, it is crucial to detect the whole structured motif directly whose spacers and other possibly significant boxes can increase its overall significance. Instead of computing a single dynamicprogramming (DP) matrix for each pair of strings, we compute β DP matrices (one for each box); and then merge the single motif occurrences of the individual boxes using the intervals of distance to determine whether they form a valid structured motif or not.
MoTeXII produces similar and partially identical results to current stateoftheart tools for structured motif extraction with respect to accuracy as quantified by statistical significance measures. Moreover, we show that it matches or outperforms these tools in terms of runtime efficiency by merging single motif occurrences efficiently. MoTeXII comes in three flavors: a standard CPU version; an OpenMPbased version; and an MPIbased version. For instance, the MPIbased version of MoTeXII requires only a couple of hours to process all human genes for structured motif extraction on 1056 processors, while current sequential tools require more than a week for this task. Finally, we show that MoTeXII is successful in extracting known composite transcription factor binding sites from real datasets.
Methods
Definitions and notation
In this section, in order to provide an overview of the algorithms used later on, we give a few definitions, generally following a standard textbook of algorithms on strings [21].
An alphabet Σ is a finite nonempty set whose elements are called letters. A string on an alphabet Σ is a finite, possibly empty, sequence of elements of Σ. The zeroletter sequence is called the empty string, and is denoted by ε. The length of a string x is defined as the length of the sequence associated with the string x, and is denoted by x. We denote by x[i], for all 1≤i≤x, the letter at index i of x. Each index i, for all 1≤i≤x, is a position in x when x≠ε. It follows that the ith letter of x is the letter at position i in x, and that x=x[1.. x].
A string x is a factor of a string y if there exist two strings u and v, such that y=u x v. Let the strings x,y,u, and v, such that y=u x v. If u=ε, then x is a prefix of y. If v=ε, then x is a suffix of y.
Let x be a nonempty string and y be a string. We say that there exists an (exact) occurrence of x in y, or, more simply, that x occurs (exactly) in y, when x is a factor of y. Every occurrence of x can be characterised by a position in y. Thus we say that x occurs at the starting position i in y when y[i.. i+x−1]=x. It is sometimes more suitable to consider the ending position i+x−1.
The edit distance, denoted by δ_{ E }(x,y), for two strings x and y is defined as the minimum total cost of operations required to transform string x into string y. For simplicity, we only count the number of edit operations and consider that the cost of each edit operation is 1. The allowed operations are the following:

Ins: insert a letter in y, not present in x; (ε,b), b≠ε;

Del: delete a letter in y, present in x; (a,ε), a≠ε;

Sub: substitute a letter in y with a letter in x; (a,b), a≠b, a,b≠ε.
The Hamming distance δ_{ H } is only defined on strings of the same length. For two strings x and y, δ_{ H }(x,y) is the number of positions in which the two strings differ, that is, have different letters. For the sake of completeness, we define δ_{ H }(x,y)=∞ for strings x, y such that x≠y.
Algorithms
In this section, we first formally define the fixedlength approximate string matching problem under the edit distance model and under the Hamming distance model; and provide a brief description and analysis of the algorithms to solve it. We show how the structured motif extraction problem can be reduced to the fixedlength approximate string matching problem, by using a stricter assumption than the one in the initial problem definition for the validity of structured motifs. Then, we provide an informal structure of our approach. Finally, we present a practical improvement on this approach by merging single motif occurrences efficiently.
Problem 1 (Edit distance)
Given a string x of length m, a string y of length n, an integer k, and an integer e<k, find all factors of y, which are at an edit distance less than, or equal to, e from every factor of fixed length k of x.
Problem 2 (Hamming distance)
Given a string x of length m, a string y of length n, an integer k, and an integer e<k, find all factors of y, which are at a Hamming distance distance less than, or equal to, e from every factor of fixed length k of x.
Let D [0.. n,0.. m] be a DP matrix, where D [i,j] contains the edit distance between some factor y[i^{′}.. i] of y, for some 1≤i^{′}≤i, and factor x[max{1,j−k+1}.. j] of x, for all 1≤i≤n, 1≤j≤m. This matrix can be obtained through a straightforward $\mathcal{O}\left(\mathit{\text{kmn}}\right)$time algorithm by constructing DP matrices D^{ s }[0.. n,0.. k], for all 1≤s≤m−k+1, where D^{ s }[i,j] is the edit distance between some factor of y ending at y[i] and the prefix of length j of x[s.. s+k−1]. We obtain D by collating D^{1} and the last row of D^{ s }, for all 2≤s≤m−k+1. We say that x[max{1,j−k+1}.. j]eoccurs in y ending at y[i]iff D[i,j]≤e, for all 1≤j≤m, 1≤i≤n.
Iliopoulos, Mouchard, and Pinzon devised MaxShift[18], an algorithm with time complexity $\mathcal{O}(m\lceil k/w\rceil n)$, where w is the size of the computer word. By using wordlevel parallelism, MaxShift can compute matrix D efficiently. The algorithm requires constant time for computing each cell D[i,j] by using wordlevel operations, assuming that k≤w. In the general case, it requires $\mathcal{O}(\lceil k/w\rceil )$ time. Hence, algorithm MaxShift requires time $\mathcal{O}\left(\mathit{\text{mn}}\right)$, under the assumption that k≤w. Notice that the space complexity is only $\mathcal{O}\left(m\right)$ since each row of D only depends on the immediately preceding row.
Theorem 1 ([18])
Given a string x of length m, a string y of length n, an integer k, and the size of the computer word w, matrix D can be computed in time $\mathcal{O}(m\lceil k/w\rceil n)$.
Let M [0.. n,0.. m] be a DP matrix, where M[i,j] contains the Hamming distance between factor y[max{1,i−k+1}.. i] of y and factor x[max{1,j−k+1}.. j] of x, for all 1≤i≤n, 1≤j≤m. Crochemore, Iliopoulos, and Pissis devised an analogous algorithm [22] that solves the analogous problem under the Hamming distance model with the same time and space complexity.
Theorem 2 ([22])
Given a string x of length m, a string y of length n, an integer k, and the size of the computer word w, matrix M can be computed in time $\mathcal{O}(m\lceil k/w\rceil n)$.
On the one hand, if the input dataset is relatively large, the possibility that there exists a structured motif which does not occur exactly, at least once, in the dataset and it also satisfies all the restrictions imposed by the input parameters, is rather unlikely, from both a combinatorial and a biological point of view. On the other hand, if the input dataset is rather small, single and structured motif extraction could potentially be performed by applying multiple sequence alignment to the input strings or exhaustive enumeration. We are therefore able to make the following stricter assumption for the validity of structured motifs.
Definition 1
A valid structured motif is called strictly valid if it occurs exactly, at least once, in any of the input strings.
Assuming that k≤w, the single motif extraction problem for strictly valid motifs can be solved in time $\mathcal{O}\left({n}^{2}\right)$ per DP matrix, where n is the average length of the N strings, thus $\mathcal{O}\left({N}^{2}{n}^{2}\right)$ in total [11]. For structured motif extraction, instead of computing a single DP matrix for each pair of strings, we compute β DP matrices (one for each box), and then merge the single motif occurrences of the individual boxes using the intervals of distance to determine whether they form a valid structured motif or not. For each pair of input strings, the DPmatrices computation requires time $\mathcal{O}\left(\beta {n}^{2}\right)$. For a pair x and y of input strings, assume the value of a cell of the first DP matrix is less than or equal to e_{1}, denoting an e_{1}occurrence of box m_{1} in y. Further, let $\delta :=max\left\{{d}_{{max}_{i}}{d}_{{min}_{i}}+1:1\le i<\beta \right\}$ and γ:=β−1. For an (e_{ i })_{1≤i≤β}occurrence of a structured motif in y, there exist $\mathcal{O}\left({\delta}^{\gamma}\right)$ possible distance sequences, each of length γ. Merging the elements of these distance sequences for x and y, for each interval separately, in a trivial way gives $\mathcal{O}\left(\gamma {\delta}^{2\gamma}\right)$ cells we have to check; thus, $\mathcal{O}\left(\gamma {\delta}^{2\gamma}{n}^{2}\right)$, in total. Combined with the time for the DPmatrices computation, in overall, the algorithm requires time $\mathcal{O}\left({N}^{2}\left(\beta +\gamma {\delta}^{2\gamma}\right){n}^{2}\right)=\mathcal{O}\left({N}^{2}\beta {\delta}^{2\gamma}{n}^{2}\right)$. In the case when each box has a fixedlength gap from the next box, that is, δ=1, the algorithm requires time $\mathcal{O}\left({N}^{2}\beta {n}^{2}\right)$.
Example 1
After the DPmatrix computation, the algorithm continues by looking for i,j≥k_{1}, such that M [i,j]≤e_{1}. The algorithm finds M [4,5]=0≤e_{1}, since δ_{ H }(x[3.. 5],y[2.. 4])=0. Let i^{′}=:i+k_{1}=7 and j^{′}=:j+k_{1}=8. In order to merge the elements of sequences s_{1} and s_{2} for a potential e_{2}occurrence of the second box, we have to check the value of δ^{2γ}=4 cells: M [i^{′}+1,j^{′}+1]; M [i^{′}+1,j^{′}+2]; M [i^{′}+2,j^{′}+1]; and M [i^{′}+2,j^{′}+2]. Only cell M [i^{′}+2,j^{′}+1]=M [9,9]=1≤e_{2}, since δ_{ H }(x[7.. 9],y[7.. 9])=1. Since q=2, AAA[1,2]TAT is a valid structured motif occurring in both CAAACCTTT and CGAAAGTAT.
A practical improvement on the runtime of the proposed algorithm can be achieved by the following observation, presented also, within a different context, in [7, 13]. The cumulative distance between two boxes distanced by ${d}_{{min}_{i}}$, from box m_{ i } to box m_{i+1}, and ${d}_{{min}_{i+1}}+1$, from box m_{i+1} to box m_{i+2}, is equivalent, from box m_{i+2} on, to the distance between boxes distanced by ${d}_{{min}_{i}}+1$, from box m_{ i } to box m_{i+1}, and ${d}_{{min}_{i+1}}$, from box m_{i+1} to box m_{i+2}. In other words, it holds that ${d}_{{min}_{i}}+\left({d}_{{min}_{i+1}}+1\right)=\left({d}_{{min}_{i}}+1\right)+{d}_{{min}_{i+1}}$. Based on this fact, limited to the ith distance interval, the prefix sums of these distance sequences form a finite arithmetic progression ${d}_{{min}_{1}}+\cdots +{d}_{{min}_{i}},\dots ,{d}_{{max}_{1}}+\cdots +{d}_{{max}_{i}}$ of length $\mathcal{O}\left(\mathrm{\delta \gamma}\right)$. Assume the value of a cell of the first DP matrix is less than or equal to e_{1}, denoting an e_{1}occurrence of box m_{1}. Merging the elements of these progressions for each interval separately gives only $\mathcal{O}\left(\gamma {\left(\mathrm{\delta \gamma}\right)}^{2}\right)$=$\mathcal{O}\left({\delta}^{2}{\gamma}^{3}\right)$ cells we have to check. Since the information for potential e_{ i }occurrences of box m_{ i }, for all 2≤i≤β, is stored in the DP matrices, we may invalidate some c>0 of the $\mathcal{O}\left({\delta}^{2\gamma}\right)$ candidates that can never yield an (e_{ i })_{1≤i≤β}occurrence in time $\mathcal{O}\left({\delta}^{2}{\gamma}^{3}+c\right)$ per e_{1}occurrence. Notice that these arithmetic progressions, and, hence, the association of the corresponding boxes with the candidates, can be precomputed, only once, since they are independent of the pairs of strings. Thus, in practice, we may avoid the enumeration of all $\mathcal{O}\left(\gamma {\delta}^{2\gamma}\right)$ DPmatrix cells. However, in the worst case, the overall time complexity of the proposed algorithm remains $\mathcal{O}\left({N}^{2}\beta {\delta}^{2\gamma}{n}^{2}\right)$.
Example 2
Let the structured motif m_{1}[1,2]m_{2}[4,5]m_{3}, where k_{1}=k_{2}=k_{3}. The arithmetic progression for the first distance interval is given by ${p}_{1}:={d}_{{min}_{1}},\dots ,{d}_{{max}_{1}}$, that is p_{1}=1,2; and for the second by ${p}_{2}:={d}_{{min}_{1}}+{d}_{{min}_{2}},\dots ,{d}_{{max}_{1}}+{d}_{{max}_{2}}$, that is p_{2}=5,6,7. Therefore by considering only p_{1}^{2}+p_{2}^{2}=13 DPmatrix cells, we may invalidate some of the δ^{2γ}=16 candidates that can never yield an (e_{ i })_{1≤i≤3}occurrence. Thus, we may avoid enumerating all γ δ^{2γ}=32 cells. This is due to the fact that this enumeration consists of only 13 distinct cells. For instance, assume M [i,j]≤e_{1}, denoting an e_{1}occurrence of box m_{1}. Let i^{′}=:i+k_{1} and j^{′}=:j+k_{1}. If cell M [i^{′}+2,j^{′}+1]>e_{2}, then we can invalidate 4 candidates. This is because the association of this cell with the 4 candidates can be precomputed.
Results
where q^{′} is the ratio (%) of q to N.
Implementation
MoTeXII was implemented in the C programming language under GNU/Linux. We implemented MoTeXII in three flavors: a standard CPU version; an OpenMP version; and an MPI version. The parallelisation scheme is beyond the scope of this article; it can be found in [11]. SMILE [23] may be used as a postanalysis programme that, given the output of a motif extractor and the input dataset, calculates the zscore and other statistical measures for assessing the statistical significance of the reported motifs. The significance of the reported motifs is computed from their occurrence frequency in a random subset of the input dataset. The support of a reported motif is defined as the total number of input sequences that contain at least one occurrence of the reported motif. The weighted support is defined as the total number of occurrences of the reported motif over all input sequences. Given the support and weighted support for each reported motif in the input dataset, SMILE computes two zscores based on the corresponding support and weighted support in the random subset. Finally, SMILE sorts the motifs by their zscores in descending order, thereby providing two ranks for each reported motif. MoTeXII can produce a SMILEcompatible output file, which can then directly be used as input for SMILE. MoTeXII is distributed under the GNU General Public License (GPL). The opensource code, the documentation, and all of the datasets referred to in this section are publicly maintained at http://www.inf.kcl.ac.uk/research/projects/motex/.
Accuracy
 1.
β single motifs m _{1},…,m _{ β } of lengths k _{1},…,k _{ β }, respectively, were generated by randomly picking k _{1}+⋯+k _{ β } letters from the DNA alphabet Σ:={A,C,G,T}.
 2.
As basic input dataset, we used N=1,062 upstream sequences of Bacillus subtilis genes of total size 240 KB, obtained from the GenBank [25] database (see [23], for details).
 3.
q (q≤N) sequences were randomly selected from these N background sequences.
 4.The following steps were performed for each of the q selected background sequences:
 (a)
An instance ${m}_{i}^{\prime}$, for all 1≤i≤β, of the single motif m _{ i } was obtained by randomly choosing e _{ i } (e _{ i }<k _{ i }) positions and randomly replacing these e _{ i } letters to one of the four letters in Σ.
 (b)
γ:=β−1 factors (spacers) g _{1},…,g _{ γ } of lengths d _{1},…,d _{ γ }, respectively, were randomly generated by randomly picking ${d}_{1}+\cdots +{d}_{\gamma}\left({d}_{{min}_{1}}\phantom{\rule{0.3em}{0ex}}\le {d}_{1}\phantom{\rule{0.3em}{0ex}}\le {d}_{{max}_{1}},\dots ,{d}_{{min}_{\gamma}}\phantom{\rule{0.3em}{0ex}}\le {d}_{\gamma}\phantom{\rule{0.3em}{0ex}}\le {d}_{{max}_{\gamma}}\phantom{\rule{0.3em}{0ex}}\right)$ letters from Σ.
 (c)
An instance ${m}^{\prime}:={m}_{1}^{\prime}{g}_{1}{m}_{2}^{\prime}{g}_{2}\dots {g}_{\gamma}{m}_{\beta}^{\prime}$ of the structured motif was generated.
 (d)
A factor r of length k _{1}+d _{1}+⋯+d _{ γ }+k _{ β } was randomly selected from the background sequence.
 (e)
Factor r was replaced by the generated instance m ^{′} of the structured motif.
 (a)
Number of motifs identified by MoTeXII using a synthetic dataset
Parameters  Implanted  Identified  Extracted 

motifs  implanted motifs  motifs  
<(8,1)[3,3](8,1),7>  100  100  100 
<(8,1)[3,3](8,1),15>  100  100  105 
<(8,1)[3,3](9,2),7>  100  100  100 
<(8,1)[3,3](9,2),15>  100  100  100 
<(9,2)[3,3](8,1),7>  100  100  128 
<(9,2)[3,3](8,1),15>  100  100  120 
<(9,2)[3,3](9,2),7>  100  100  101 
<(9,2)[3,3](9,2),15>  100  100  100 
Statistical evaluation of motifs identified by MoTeXII using a synthetic dataset
Parameters  Implanted  Identified  Extracted  Ranking of 

motifs  implanted motifs  motifs  implanted motif  
<(3,0)[2,2](5,0),7>  1  1  5  1/1 
<(5,0)[2,2](3,0),7>  1  1  6  1/1 
<(3,0)[2,2](6,1),7>  1  1  2,475  1/1 
<(6,1)[2,2](3,0),7>  1  1  2,753  1/1 
<(5,1)[2,2](6,1),7>  1  1  17,118  1/1 
<(6,1)[2,2](5,1),7>  1  1  17,135  1/1 
Efficiency
To evaluate the efficiency of MoTeXII, we compared its performance to the corresponding performance of RISOTTO and EXMOTIF, which are currently the most widelyused tools for structured motif extraction.
Elapsedtime comparison of RISOTTO, EXMOTIF, and MoTeXII using a smallscale real dataset
Parameters  RISOTTO  EXMOTIF  MoTeXIICPU  MoTeXIIOMP t 48 

<(8,1)[2,3](8,1),7>  286s  898s  1,885s  46s 
<(8,1)[2,3](8,1),15>  217s  626s  1,860s  48s 
<(8,1)[2,3](9,2),7>  2,086s  2,253s  1,871s  49s 
<(8,1)[2,3](9,2),15>  1,103s  2,222s  1,860s  48s 
<(9,2)[2,3](8,1),7>  4,868s  2,222s  1,868s  48s 
<(9,2)[2,3](8,1),15>  4,279s  2,197s  1,856s  49s 
<(9,2)[2,3](9,2),7>  39,488s  22,862s  1,871s  47s 
<(9,2)[2,3](9,2),15>  21,274s  22,739s  1,865s  47s 
Elapsedtime comparison of RISOTTO, EXMOTIF, and MoTeXII using a mediumscale real dataset
Parameters  RISOTTO  EXMOTIF  MoTeXIIOMP 

t 48  
<(8,1)[3,5](8,1),10>  1,015s  **  6,853s 
<(8,1)[3,5](8,1),20>  423s  **  6,848s 
<(8,1)[3,5](10,3),10>  *  **  6,865s 
<(8,1)[3,5](10,3),20>  41,310s  **  6,915s 
<(10,3)[3,5](8,1),10>  492,282s  **  7,002s 
<(10,3)[3,5](8,1),20>  *  **  6,976s 
<(10,3)[3,5](10,3),10>  *  **  7,008s 
<(10,3)[3,5](10,3),20>  *  **  7,005s 
Elapsedtime comparison of RISOTTO, EXMOTIF, and MoTeXII using a largescale real dataset
Parameters  RISOTTO  EXMOTIF  MoTeXIIMPI np 1056 

<(8,1)[2,3](9,2)[3,5](10,3),5>  *  *  12,068s 
<(8,1)[2,3](10,3)[3,5](9,2),5>  *  *  12,371s 
<(9,2)[2,3](8,1)[3,5](10,3),5>  *  *  11,953s 
<(9,2)[2,3](10,3)[3,5](8,1),5>  *  *  12,095s 
<(10,3)[2,3](8,1)[3,5](9,2),5>  *  *  12,035s 
<(10,3)[2,3](9,2)[3,5](8,1),5>  *  *  11,729s 
Real applications
To further evaluate the accuracy of MoTeXII in extracting known composite transcription factor binding sites from real datasets, we compared its output to the corresponding output of EXMOTIF using SMILE.
Extraction of transcription factors for the Zinc factors by EXMOTIF and MoTeXII
EXMOTIF  MoTeXII  

TF name  Known motif  Predicted Motif  Extracted motifs  Ranking  Extracted motifs  Ranking 
GAL4  
GAL4 chips  CGGRnnRCYnYnCnCCG  CGG[11,11]CCG  1634(3346)  1/1  1634(3346)  1/1 
CAT8  CGGnnnnnnGGA  CGG[6,6]GGA  1621(3356)  451/73  1621(3356)  359/51 
HAP1  CGGnnnTAnCGGCGGnnnTAnCGGnnnTA  CGG[6,6]CGG  1621(3356)  84/96  1621(3356)  73/85 
LEU3  RCCGGnnCCGGY  CCG[4,4]CGG  1588(3366)  2/2  1588(3366)  1/2 
LYS  WWWTCCRnYGGAWWW  TCC[3,3]GGA  1605(3360)  39/25  1605(3360)  32/17 
PPR1  WYCGGnnWWYKCCGAW  CGG[6,6]CCG  1621(3356)  1/2  1621(3356)  1/2 
PUT3  YCGGnAnGCGnAnnnCCGA  
CGGnAnGCnAnnnCCGA  CGG[10,11]CCG  727(4035)  1/1  727(4035)  1/1 
with rank 290 out of 5371 final valid motifs and a zscore of 22.61. As a direct comparison, identical results were reported by EXMOTIF.
Conclusions and discussion
In this article, we introduced MoTeXII, a wordbased HPC tool for both single and structured MoTif eXtraction from largescale datasets. A valid structured motif is called strictly valid if it occurs exactly, at least once, in any of the input sequences. By making this stricter assumption for motif validity, we showed how the structured motif extraction problem can be reduced to the fixedlength approximate string matching problem. Surprisingly, this natural and simple reduction has never been considered in the literature.
As a direct result of this reduction, and assuming that the length of every single motif is less than or equal to the size of the computer word, the runtime of MoTeXII does not depend on (i) the length for motifs, (ii) the size of the alphabet, or (iii) the error thresholds. Moreover, MoTeXII is guaranteed to find globally optimal solutions. It can identify structured motifs under the edit distance model or the Hamming distance model. Finally, MoTeXII also comes in two HPC flavors: the OpenMPbased version and the MPIbased version.
Stateoftheart wordbased motif extractors produce globally optimal solutions but exhibit many disadvantages. We demonstrated that MoTeXII can alleviate these shortcomings for structured motif extraction from small, medium, and largescale datasets. The scalability of our approach is due to the fact that the proposed algorithm is independent of the aforementioned input parameters and is highly parallelisable. For instance, we showed how the quadratic time complexity of MoTeXII can be slashed, in theory and in practice, by using parallel computations; whereas suffixtreebased motif extractors are difficult to parallelise effectively. The extensive experimental results presented are promising, both in terms of accuracy under statistical measures of significance as well as efficiency; a fact that suggests that further maintenance and development of MoTeXII is desirable.
For future work, we will explore the possibility of optimising our approach by using lossless filters (see [19] and [20], for instance) for eliminating a possibly large fraction of the input that is guaranteed not to contain any valid occurrence before completing the motif inference task. Our main goal is to accurately detect single and structured motifs over massive sets of biological sequences representing a set of species. We are especially interested in discovering transcription factor binding sites whose conservation is decreasing as the evolutionary distance between those species increases. We plan to employ MoTeXII in a phylogenetic framework to incorporate evolutionary information in the motif extraction process.
Availability and requirements

Project name: MoTeX

Project home page: http://www.inf.kcl.ac.uk/research/projects/motex/

Operating system: GNU/Linux

Programming language: C

Other requirements: gcc version 4.6.3 or higher

License: GNU GPL

Any restrictions to use by nonacademics: licence needed
Declarations
Acknowledgements
The publication costs for this article were partially funded by the Department of Informatics at King’s College London. This work was partially supported by a Research Grant (#RG130720) awarded by the Royal Society. We thank Stilianos Arhondakis (Enzyme Technology & Genomics laboratory) from the Institute of Molecular Biology and Biotechnology (IMBB) of the Foundation for Research and Technology – Hellas (FORTH) for valuable comments and useful discussions.
Authors’ Affiliations
References
 Applied Combinatorics on Words. Edited by: Lothaire M. 2005, Cambridge, UK: Cambridge University PressGoogle Scholar
 Pavesi G, Mereghetti P, Mauri G, Pesole G: Weeder web: discovery of transcription factor binding sites in a set of sequences from coregulated genes. Nucleic Acids Res. 2004, 32 (WebServerIssue): 199203.View ArticleGoogle Scholar
 Rombauts S, Déhais P, Van Montagu M, Rouzé P: PlantCARE, a plant cisacting regulatory element database. Nucleic Acids Res. 1999, 27 (1): 295296. 10.1093/nar/27.1.295.View ArticlePubMed CentralPubMedGoogle Scholar
 van Helden J, Andre B, Vides CJ: Extracting regulatory sites from the upstream region of yeast genes by computational analysis of oligonucleotide frequencies. J Mol Biol. 1998, 281 (5): 827842. 10.1006/jmbi.1998.1947.View ArticlePubMedGoogle Scholar
 Eskin E, Pevzner PA: Finding composite regulatory patterns in dna sequences. Bioinformatics. 2002, 18 (Suppl 1): 354363. 10.1093/bioinformatics/18.suppl_1.S354.View ArticleGoogle Scholar
 Sagot MF: Spelling approximate repeated or common motifs using a suffix tree. Proceedings of the 3rd Latin American Symposium on Theoretical Informatics (LATIN’98). 1998, London, UK: Springer, 374390.Google Scholar
 Carvalho AM, Freitas AT, Oliveira AL, Sagot MF: An efficient algorithm for the identification of structured motifs in DNA promoter sequences. IEEE/ACM Trans Comput Biol Bioinformatics. 2006, 3 (2): 126140. 10.1109/TCBB.2006.16.View ArticleGoogle Scholar
 Das M, Dai HK: A survey of DNA motif finding algorithms. BMC Bioinformatics. 2007, 8 (Suppl 7): 2110.1186/147121058S7S21.View ArticleGoogle Scholar
 Sinha S, Tompa M: YMF: A program for discovery of novel transcription factor binding sites by statistical overrepresentation. Nucleic Acids Res. 2003, 31 (13): 35863588. 10.1093/nar/gkg618.View ArticlePubMed CentralPubMedGoogle Scholar
 Floratou A, Tata S, Patel JM: Efficient and accurate discovery of patterns in sequence data sets. Knowl Data Eng IEEE Trans. 2011, 23 (8): 11541168.View ArticleGoogle Scholar
 Pissis SP, Stamatakis A, Pavlidis P: MoTeX: A wordbased HPC tool for MoTif eXtraction. Fourth ACM International Conference on Bioinformatics and Computational Biology (ACMBCB 2013). Edited by: ACM. 2013, New York, NY, USA: ACM, 1322.Google Scholar
 Carvalho AM, Marsan L, Pisanti N, Sagot MF: RISOTTO: fast extraction of motifs with mismatches. Proceedings of the 7th Latin American Symposium on Theoretical Informatics (LATIN’06), Valdivia, Chile. 2006, Berlin Heidelberg: Springer, 757768.Google Scholar
 Carvalho AM, Freitas AT, Oliveira AL, Sagot MF: A highly scalable algorithm for the extraction of cisregulatory regions. Proceedings of the 3rd Asia Pacific Bioinformatics Conference Advances in Bioinformatics and Computational Biology. Edited by: Chen YPP, Wong L. 2005, Singapore: Imperial College Press, 273282.View ArticleGoogle Scholar
 Zhang Y, Zaki M: EXMOTIF: efficient structured motif extraction. Algo Mol Biol. 2006, 1 (1): 118. 10.1186/1748718811.View ArticleGoogle Scholar
 Na JC, Apostolico A, Iliopoulos CS, Park K: Truncated suffix trees and their application to data compression.Theor Comput Sci. 2003, 304 (13): 87101. 10.1016/S03043975(03)000537.View ArticleGoogle Scholar
 Jia C, Carson M, Yu J: A fast weak motiffinding algorithm based on community detection in graphs. BMC Bioinformatics. 2013, 14 (1): 114. 10.1186/14712105141.View ArticleGoogle Scholar
 Carvalho AM, Oliveira AL, Freitas AT, Sagot MF: A parallel algorithm for the extraction of structured motifs. Proceedings of the 2004 ACM Symposium on Applied Computing. SAC ’04. 2004, Nicosia, Cyprus: ACM, 147153.View ArticleGoogle Scholar
 Iliopoulos C, Mouchard L, Pinzon Y: The, MaxShift algorithm for approximate string matching. Proceedings of the Fifth International Workshop on Algorithm Engineering (WAE 2001). Lecture Notes in Computer Science. Edited by: Brodal G, Frigioni D, MarchettiSpaccamela A. 2001, Denmark: Springer, 1325.Google Scholar
 Federico M, Peterlongo P, Pisanti N, Sagot MF: Finding long and multiple repeats with edit distance. Proceedings of the Prague Stringology Conference 2011. Edited by: Holub J, žďárek J. 2011, Czech Republic: Czech Technical University in Prague, 8397.Google Scholar
 Federico M, Peterlongo P, Pisanti N, Sagot MF: RIME: Repeat identification. Discrete Appl Math. 2014, 163 Part 3 (0): 275286.View ArticleGoogle Scholar
 Crochemore M, Hancart C, Lecroq T: Algorithms on Strings. New York. 2007, USA: Cambridge University PressView ArticleGoogle Scholar
 Crochemore M, Iliopoulos CS, Pissis SP: A parallel algorithm for fixedlength approximate stringmatching withkmismatches. Algorithms and Applications. Lecture Notes in Computer Science. Edited by: Elomaa T, Mannila H, Orponen P. 2010, Berlin Heidelberg: Springer, 92101.Google Scholar
 Marsan L, Sagot MF: Algorithms for extracting structured motifs using a suffix tree with an application to promoter and regulatory site consensus identification. J Comput Biol: J Comput Mol Cell Biol. 2000, 7 (34): 345362. 10.1089/106652700750050826.View ArticleGoogle Scholar
 Buhler J, Tompa M: Finding motifs using random projections. J Comput Biology: J Comput Mol Cell Biol. 2002, 9 (2): 225242. 10.1089/10665270252935430.View ArticleGoogle Scholar
 GenBank. [http://www.ncbi.nlm.nih.gov/genbank/],
 Ensembl Genome Browser. [http://www.ensembl.org/index.html],
 Helden Jv, Rios AF: ColladoVides J: Discovering regulatory elements in noncoding sequences by analysis of spaced dyads. Nucleic Acids Res. 2000, 28 (8): 18081818. 10.1093/nar/28.8.1808.View ArticlePubMed CentralPubMedGoogle Scholar
 GuhaThakurta D, Stormo GD: Identifying target sites for cooperatively binding factors. Bioinformatics. 2001, 17 (7): 608621. 10.1093/bioinformatics/17.7.608.View ArticlePubMedGoogle Scholar
 Zhu J, Zhang MQ: SCPD: A promoter database of the Yeast Saccharomyces cerevisiae. Bioinformatics. 1999, 15 (7): 607611. 10.1093/bioinformatics/15.7.607.View ArticlePubMedGoogle Scholar
Copyright
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly credited. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.