logo

FAQ: Frequently asked questions about this server.

1. What is DNA motifs?

A DNA motif is a region of DNA that regulates the expression of downstream genes located on that same molecule of DNA, i.e. a chromosome. This concept can be equivalent to DNA cis-regulatory element or cis-element. It contains the transcription factor binding sites (TFBSs) and other conserved functional elements in the 5 intergenic regions of genes. We use motifs instead of DNA motifs in the following.

2. What can your server do in motif analyses?

Our server has a number of novel capabilities (i) de-novo motif finding; (ii) motif refinement and evaluation based on information extracted from the entire genome and a phylogenetic footprinting method, (iii) motif scanning based on a global P-value estimation method, (iv) motif comparison and clustering using a novel and effective technique and (v) analysis of motifs co-occurrences in the regulatory regions.

3. How does DOOR2 database support this server?

(http://csbl.bmb.uga.edu/DOOR/) is a complete and reliable operon database covering 2,072 bacteria genomes and with overall accuracy of ~90% evaluated by Brouwer (2008) on Brief Bioinformatics. It is currently integrated with our motif analyses server. Specifically, for motif analyses in prokaryotic genomes, the users do not need to provide the operons and corresponding promoters information themselves; they can go to our server only with their questions and click several buttons for analyzing the motifs.

4. What kinds of formats of input DNA sequences are acceptable for this server?

FASTA is the only acceptable format.
See details in (http://en.wikipedia.org/wiki/FASTA_format)

5. What kinds of formats of motifs are acceptable as input for this server?

Currently we accept three kinds of motif formats, shown as follows,
Motif alignment, e.g.:
>ECK120011345
AACATTTAGTTAACC
TAAAAATTGTTAACA
AAAACTTGATTAACA
AACATTTAGTTAACT
AACAATTATTTAACA
TAATTATTATTAACC
AAAATATAATGAACA

Motif matrix, e.g.:
>ECK120011345

A 40 47 23 42 23 33 12 23 40

G 5 6 8 9 5 15 0 13 26

C 7 5 30 7 14 1 5 14 0

T 23 17 14 17 33 26 58 25 9


Motif consensus, e.g.:
>consensus1
GCTTTTGATGACTTCAAACAC
W = A or T
S = C or G
R = A or G
Y = C or T
K = G or T
M = A or C
B = C, G, or T (not A)
D = A, G, or T (not C)
H = A, C, or T (not G)
V = A, C, or G (not T)
N = A, C, G, or T

6. What is position weight matrix (PWM)?

A position weight matrix (PWM), also known as a position-specific weight matrix (PSWM), is a commonly used representation of motifs (patterns) in biological sequences. PWMs are often derived from a set of aligned sequences that are thought to be functionally related.
A PWM has one row for each symbol of the alphabet: 4 rows for nucleotides in DNA sequences. It also has one column for each position in the pattern. A basic PWM using relative frequencies is constructed by counting the occurrences of each symbol at each position and then normalizing at each position. Formally, given a set X of N aligned sequences of length l, the elements of the PWM M are calculated:

Drawing
where i (1,...,N), j (1,...,l), k is the set of symbols in the alphabet and I(a=k) is an indicator function where I(a=k) is 1 if a=k and 0 otherwise.

7. What is a Position-Specific Scoring Matrix (PSSM)

Often the element in PWMs are calculated as log likelihoods. That is, the elements of the PWM are transformed using a background model so that:

Drawing
The simplest background model assumes that each letter appears equally frequently in the dataset. That is, the value of for all symbols in the alphabet (0.25 for nucleotides and 0.05 for amino acids). Applying this transformation to the PWM M from above gives:


Drawing

The entries -inf in the matrix make clear the advantage of adding pseudo counts, especially when using small datasets to construct M. The background model need not have equal values for each symbol: for example, when studying organisms with a high GC-content, the values for C and G may be increased with a corresponding decrease for the A and T values.

8. What is the information content of a PWM?(PSSM)

The information content (IC) of a PWM is sometimes of interest, as it says something about how different a given PWM is from a uniform distribution.
The self-information of observing a particular symbol at a particular position of the motif is:

Drawing
The expected (average) self-information of a particular element in the PWM is then:

Drawing
Finally, the IC of the PWM is then the sum of the expected self-information of every element:

Drawing
Often, it is more useful to calculate the information content with the background letter frequencies of the sequences you are studying rather than assuming equal probabilities of each letter (e.g., the GC-content of DNA of thermophilic bacteria range from 65.3 to 70.8, thus a motif of ATAT would contain much more information than a motif of CCGG). The equation for information content thus becomes

Drawing
where pb is the background frequency for that letter. This corresponds to the Kullback-Leibler divergence or relative entropy. However, it has been shown that when using PSSM to search genomic sequences this uniform correction can lead to overestimation of the importance of the different bases in a motif, due to the uneven distribution of n-mers in real genomes, leading to a significantly larger number of false positives.
(http://en.wikipedia.org/wiki/Position_weight_matrix)

9. How do you de-novo identify motifs?

The de-novo motif finding program, BOBRO, is published on NAR, 2011.
Guojun Li, Bingqiang Liu, Qin Ma, Ying Xu,A new framework for identifying cis-regulatory motifs in prokaryotes,Nucleic Acids Res.2011 Apr;39(7):e42
It is an algorithm for prediction of cis regulatory motifs in a given set of promoter sequences. The algorithm substantially improves the prediction accuracy and extends the scope of applicability of the existing programs based on two key new ideas: (i) a highly effective method for reliably assessing the possibility for each position in a given promoter to be the (approximate) start of a conserved sequence motif; and (ii) a highly reliable way for recognition of actual motifs from the accidental ones based on the concept of motif closure. These two key ideas are embedded in a classical framework for motif finding through finding cliques in a graph but have made this framework substantially more sensitive as well as more selective in motif finding in a very noisy background. A comparative analysis shows that the performance coefficient was improved from 29% to 41% by our program compared to the best among other six state-of-the-art prediction tools on a large-scale data sets of promoters from one genome, and also consistently improved by substantial margins on another kind of large-scale data sets of orthologous promoters across multiple genomes. The power of BOBRO in dealing with noisy data was further demonstrated through identification of the motifs of the global transcriptional regulators by running it over 2390 promoter sequences of Escherichia coli K12. The related data sets and results can be found at: [http://csbl.bmb.uga.edu/~maqin/motif_finding/](http://csbl.bmb.uga.edu/~maqin/motif_finding/).

10. How do you refine the predicted motifs?

The motif refinement and evaluation function is so called BBR (BoBro-based motif refinement), a method for filtering out noises among predicted motifs at a genome scale. Consider a genome-scale motif prediction problem: denote all the motifs predicted by a de-novo?motif finding tool as Drawing; R and C represent the given set of regulatory sequences for motif identification and a control sequence set, respectively. For any motif Drawing , it is considered as a motif if it satisfies the following three criteria: (i) the P-value of m with respect to a hypothesis that it appears in R by chance is below a specified cutoff value; (ii) R is more enriched of the instances of m than C, as defined in formula (1); and (iii)?m?is well-conserved across a diverse set of species, as defined in formula (2).
Criterion (i) is measured using the P-value defined in our previous work (Liet al., 2011). Specifically, let x be a random variable denoting the number of instances of a motif in a given set of regulatory sequences, and its probability distribution,p(x), can be approximated using a Poisson distribution. Hence, the P-value of a motif can be calculated by summing up the probability of p(x)Drawing over, denoting that the motif has at least k instances. An enrichment score is defined to evaluate the statistical significance of the ratio between the number of m instances in R and that in C, as given in the following,
Drawing(1)
where NR and NC are the numbers of instances of m in R and C, respectively; and |R| and |C| are the sequence lengths of R and C, respectively. Criterion (iii) is defined in terms of the average enrichment score defined in formula (2), with each?Z?term being defined in (1) for each organism over a set of diverse species and the original genome.
Drawing(2)
where Drawing represents a set of species and Drawing is m enrichment score in species i,Drawing. We consider a motif as statistically significant if its P < 3.3e-5 (the P-value threshold has been corrected for multiple testing based on the estimated number, 300, of TFs in?E. coli). Criteria (ii) and (iii) are designed to ensure that predicted motifs will be as biologically meaningful as possible (Bailey, 2011).

11. How do you do motif scanning for additional motif instances across a genome based on known or predicted motifs?

The motif scanning function is so-called BBS (BoBro-based Scanning), scanning and ranking new instances of a query motif based on P-values. A key to reliable motif-scanning at a genome scale is an ability to effectively evaluate the similarity between a motif instance and a query motif (Das and Dai, 2007; Haverty and Weng, 2004;Medina-Rivera et al., 2011; Thomas-Chollier et al., 2008). Obviously, different similarity cutoffs may result in different scanning results. BBS provides a global P-value for the entire motif instances for each motif scan. We first introduce a few definitions; let M be an aligned query motif of L nucleotides long and its PWMDrawing is defined as a 4-by-L matrix, given in (3):
Drawing(3)
where Drawingis the probability of nucleotide Drawing appearing at position j in M; and Drawingis the probability of i appearing in the background sequences, e.g. all the promoter sequences in the entire genome. Comparing with the traditional PWM model that assumes independence among different sequence positions, we assumed first-order Markov-chain property among consecutive sequence positions in our model. We generated a transition matrixDrawing, with Drawing representing the probability of a specific nucleotide type i followed by a specific nucleotide type i in consecutive positions j and j + 1 of the query motif. The similarity between a motif instance
DrawingDrawing
and a query motif M is measured using:
Drawing(4)
Consider a motif Drawing with t instances Drawing, the average similarity Drawingbetween Drawing and M is measured using the following:
Drawing(5)
A Drawing closure of M, denoted as Drawing, is a set of sequence segments in the input regulatory sequences, each having a similarity score no less than Drawing. Our previous experience has been that the documented cis-regulatory motifs tend to have significantly more instances with high similarities among them than the accidental ones, and the size of a Drawingclosure provides a good measure for this (Li?et al., 2011). The P-value Drawing of Drawing can be approximated using a Poisson distribution based on our previous work (Li et al., 2011). We can select a DrawingvalueDrawingso that theDrawingclosure of M can give the best motif prediction performance measured in terms of prediction sensitivity and specificity. One way to accomplish this is through finding a Drawingthat minimizes the following function:
Drawing(6)
This capability can be used to derive an optimal similarity cut-off for motif scanning on a statistically sound basis.

12. What is background genome?

A background genome used in our web server represents an entire genome or a set of control regulatory sequences which are known not to contain query motifs. Such a background genome can be used to refine and evaluate the predicted motif by generating a Z-score.

13. How do you do motif comparison between predicted motifs and annotated motifs in motif databases?

The motif comparison function utilize weak conserved signals of motifs flanking regions in motif comparison: We have observed that the flanking regions of cis-regulatory motifs tend to have some level of sequence conservation, and we have developed the following procedure to take advantage of this information in motif comparison. Define a deformation of?information content (Schneider?et al., 1986) for a motif?M?of length?L?as follows:
Drawing(7)
whereDrawingand the other items are the same as in formula (3). Consider two motifs M1 and M2 with lengths L1 and L2, respectively, and L= min {L1,L2}. Let M1 and M2 be the two extended motifs formed by concatenating theDrawingandDrawingnucleotides on each side of each motif instance sequence of M1 and M2, respectively (If the location information of given motifs in their original genome is available, we can use the flanking region of each motif to generate the extended motif sequence); hence, their lengths are 2L1 and 2L2. The similarity between the extended instances of M1 and M2 is defined as follows:
Drawing(8)
Where
Drawing(9)
And
. Drawing(10)

14. What is the theoretical complexity of the back-end algorithms and acutal computational capability of DMINDA?

Table 1. The theoretical complexity, real computation time and parameters of related tools; L and n means motif length and the number of motif instances, M and N means the size of input promoter sequences and control sequences (number of nucleic acids), P means the number of promoters involved, t is the number of simulations for calculation of the p-values of motif closures. All the programs are implemented on a computer with 264GB memory and CPU E5-2630 0 @ 2.3 GHz.

Program Theoretical time complex Input Real User Sys Parameter
BOBRO O(M2L2)+O(tML) 2390 length 300nt promoters 1825m 1823m 1m19s -k 5 -c 1.00 -u 0.70 -e 3 -w 2.00 -b 0.95 -N 5 -l 14 -F -o 500
BOBRO O(M2L2)+O(tML) 2390 Intergenetic regions 2181m 2178m 1m2s -k 5 -c 1.00 -u 0.70 -e 3 -w 2.00 -b 0.95 -N 5 -l 14 -F -o 500
BBS O(Ln+LM) 245 Human motif and promoter 161m 160m 0m41s -w
BBR O(LM+LN) ~300 motifs ,271 reference genomes 2311m 2254m 27m --
BBC Comparison/Clustring : O(Ln) /O(n2logn) Pairwise Comparison 561 motifs 1m 0m22s 0m14s Clustering: T1: 0.85; T2:0.91
BBA O(Pn) 159 E. coli motifs pair-wise analysis (12561 pairs) 41m57.215s 21m53.203s 10m22.731s --

Table 2. The actual computational time of samples and some large-scale jobs on DMINDA. Note: The number of output motifs should be less than 100, otherwise they will be too slow to be displayed.

BoBro BBS BBC BBA
Sample data JobID: 20140316135117f
Input: 19 promoters
Output: 8 motifs
Time: 120s
JobID: 20140316133439s
Input: 5 motifs and 19 promoters
Time: 50s
JobID: 20140316133454c
Input: 5 motifs
Time: 9s
Input: 8 motifs and
19 promoters
Time:6s
A biological pathway: TCA cycle JobID: 20140120153137f
Input: 17 promoters Output: 10 motifs
Time: 238s
JobID: 2014031691048s
Input: 10 motifs and 17 promoters
Time: 74s
JobID: 2014031691441c
Input: 10 motifs and 17 promoters
Time:16s
Input: 17 promoters and 10 motifs
Time: 8s
A Bacterial genome: NC_012034 JobID: 20140316125905f
Input: 1,272 promoters
Output: 80 motifs
Time: 6,908s
Job ID: 20140316151804s
Input: 80 motifs and 1,272 promoters
Time: 448s
JobID: 20140316151926c
Input: 80 motifs and 1,272 promoters
Time: 20s
Input: 80 motifs and 1272 promoters
Time: 527s
The Human genome N/A JobID: 20140316101840s
Input: 5 motifs and 20,044 promoters
Time: 447s
JobID: 20140316103154c
Input: 5 motifs and 20,044 promoters
Time: 13s
Input: 5 motifs and 20,044 promoters
Time: 4s
Limit of to-be-shown motifs 100 100 100 100

15. How do you do motif clustering in your server?

Motif clustering using the new similarity measure: A group of motifs can be clustered into subgroups of similar motifs using the following algorithm, which is based on a maximum spanning tree (MST) representation of the candidate motifs. First, consider a complete graph defined over a list of candidate motifs, each represented as a node and each pair of motifs connected by an edge; the weight of an edge is the similarity between the two corresponding motifs (see Fig. 1a). An MST of the graph is constructed using Kruskals algorithm (Thomas, 2001). We have clustered the predicted motifs based on two different similarity thresholds, T1 and T2, giving rise to two classes of motif clusters, namely, highly reliable and relatively reliable motif clusters, respectively. We have compared each pair of documented motifs in the RegulonDB database (Salgado et al., 2013) and assigned the median and the upper quartile of all the similarities to T1 and T2, respectively. Each of the two thresholds is used to remove edges with similarities lower than the threshold, giving rise to the final list of motif clusters (see Fig. 1) represented as a connected sub-tree of the MST after application of the threshold. Then, all instances of each motif cluster are mapped back to the original regulatory sequences, facilitating further analysis and interpretation of the motif-prediction results.
Drawing
Figure 1: An example of a two-level clustering of motifs using a minimal spanning tree, consisting of six motifs: (a) a complete similarity graph is constructed with the weight of each edge representing the two corresponding motifs similarity; (b) an MST {(1,2), (2,6), (3,6), (3,4), (5,6)} is constructed using Kruskals algorithm; (c) four connected components of the MST created using the first-level threshold , i.e. {1, 2}, {5, 6}, {3} and {4}, reflecting that motifs 1 and 2 and 5 and 6 are similar compared with the other motif pairs; and (d) the motif cluster {1, 2} is split into two dependent motif clusters {1} and {2} using the threshold , reflecting that the similarity between motifs 1 and 2 are lower than that between motifs 5 and 6.

16. How do you do motif co-occurrence analysis?

We have implemented a function BBA to evaluate the co-occurrences among the identified motifs in a given set of regulatory sequences, which can reveal joint regulation relationships by multiple TFs. For a given motif pair?a?and?b, and the entire set of promoter sequences P, let A and B be the subsets of P that contain motif instances of a and b, respectively (we assume, without loss of generality,Drawing). Let k=Drawing, then the probability of A and B sharing k promoter sequences can be calculated using the following hyper-geometric function:
Drawing(11)
The P-value of a and b co-occurring in the same regulatory regions is calculated as the probability of A and B sharing at least k regulatory sequences. For a pair of motifs, a significant P-value means their instances tend to occur in same regulatory sequences, hence indicating that their corresponding TFs may co-regulate the same genes with high probability.

17. Where can I get the algorithm parameters for my submitted job?

All the algorithm parameters can be found in the top of the results which are ready for download on the right-upper corner.

18. Where can we download the source codes of your programs along with their documentations?

Please go to (http://csbl.bmb.uga.edu/DMINDA/download.php) for the source codes along with their documentations.

19. What are the performances of your motif analyses functions compared to other programs?

We have carried out systematic comparisons between motif predictions by BoBro2.0 and by the MEME package. The comparison results on E. coli K12 genome and the Human genome show that BoBro2.0 can identify the statistically significant motifs at a genome scale more efficiently, identify motif instances more accurately and get more reliable motif clusters than MEME. In addition BoBro2.0 provides correlational analyses among the identified motifs to facilitate the inference of joint regulation relationships of transcription factors. More details can be found in the following reference:
Qin Ma, Bingqiang Liu, Chuan Zhou, Yanbin Yin, Guojun Li, Ying Xu, BoBro2.0: An integrated toolkit for accurate prediction and analysis of cis regulatory elements at a genome scale.Bioinformatics,10.1093/bioinformatics/btt397, 2013 (PMID:23846744)

20. Do I have to leave my email address for retrieve the results?

No, the email address is optional and you can record your unique job ID and retrieve your results by the searching bar, appearing on every page in our server.

21. How to cite your papers?

Please cite the following two papers if you use our server: (http://csbl.bmb.uga.edu/DMINDA/index.php#tabs-3)
(i) Qin Ma*, Hanyuan Zhang*, Xizeng Mao, Chuan Zhou, Bingqiang Liu, Xin Chen, Ying Xu, An integrated high-performance web server for DNA motif analyses, submitted to Nucleic Acids Research (2014 Web Server Issue).
(ii) Qin Ma, Bingqiang Liu, Chuan Zhou, Yanbin Yin, Guojun Li, Ying Xu, BoBro2.0: An integrated toolkit for accurate prediction and analysis of cis regulatory elements at a genome scale.Bioinformatics,10.1093/bioinformatics/btt397, 2013 (PMID:23846744)
(iii) Guojun Li, Bingqiang Liu, Qin Ma, Ying Xu, A new framework for identifying cis-regulatory motifs in prokaryotes, Nucleic Acids Research. 2011 Apr;39(7):e42(PMID:21149261)