Home      Labs      Publications      People      Tools   

From CAGT

Clover: Cis-eLement OVERrepresentation

Introduction

Clover is a program for identifying functional sites in DNA sequences. If you give it a set of DNA sequences that share a common function, it will compare them to a library of sequence motifs (e.g. transcription factor binding patterns), and identify which if any of the motifs are statistically overrepresented in the sequence set.

Publication

Martin C Frith, Yutao Fu, Liqun Yu, Jiang-Fan Chen, Ulla Hansen, Zhiping Weng (2004). Detection of functional DNA motifs via statistical over-representation. Nucleic Acids Research 32(4):1372-81.
Here are the data sets studied in the paper.

Changes

2-21-2005: Made the source code slightly more portable.
2-16-2005: Fixed the source code for fussy compilers.
3-3-2004: Fixed crashes on some kinds of bad input.
10-26-2003: Added minor error checking. SGI version available.

Get Clover

  1. Download Clover by clicking on one of these links, and saving the file on your computer:
    Clover executable for Linux (RedHat 7.2/7.3)
    Clover executable for Sun (Solaris 8)
    Clover executable for SGI/IRIX
    Clover executable for Mac OS X (Oct 26 2003 version)
  2. Set execute permission for the file by typing chmod +x clover-linux (or whatever you saved it as).
  3. Clover is now ready to run.
OR
  1. Download the Clover source code.
  2. Uncompress: gunzip clover-src.tar.gz
  3. Un-archive: tar -xvf clover-src.tar
  4. Change directory: cd clover-src
  5. Compile (cross your fingers): make

Unfortunately the source code doesn't compile successfully on all systems. We'd love to hear your suggestions for making it more portable.

Get a Motif Library

To use Clover, you will need a library of sequence motifs. We provide the JASPAR collection of transcription factor binding site patterns. JASPAR is described in the following publication; please give suitable credit:
Lenhard B, Sandelin A, Mendoza L, Engstrom P, Jareborg N, Wasserman WW. Identification of conserved regulatory elements by comparative genome analysis. J Biol 2003 2(2):13.

Another source of motifs is TRANSFAC: the commercial nature of this database prevents us from providing it directly.

Get Background Sequences

We recommend using background sequence sets to contrast with your sequences of interest. Which background sets to use depends on which sequences you are studying: e.g. they should come from the same taxonomic group and have similar GC compositions. Here are some background sets we have used:

These files need to be uncompressed (using gunzip) before using them with Clover.

Required Input

Clover requires two inputs: a file of DNA sequences in FASTA format (here is an example), and a file of sequence motifs in a FASTA-like format described below. Any non-alphabetic characters in the sequences are ignored, and any alphabetic characters except A, C, G, T (uppercase or lowercase) are converted to 'n' and excluded from matching motifs. The motif file should look like this:
>TATA
0 0 0 10
10 0 0 0
0 0 0 10
10 0 0 0
>E-box
1 20 1 1
(etc)

Each motif begins with a title line containing the character '>' followed by the motif's name. Subsequent lines represent successive positions of the motif, from 5' to 3', and the columns contain counts of A, C, G, and T, respectively, observed at each position. These numbers typically come from an alignment of several binding sites for a transcription factor.

Raw Scores and P-values

Clover will compare each motif in turn to the sequence set, and calculate a "raw score" indicating how strongly the motif is present in the sequence set. Raw scores by themselves are hard to interpret, so Clover provides options (which we recommend you use) to determine the statistical significance of the raw scores. Four ways of determining statistical significance are available. The first involves providing Clover with one or more files of background DNA sequences. Each background file should contain sequences in FASTA format, with total length much greater than the target sequence set. For each background set, Clover will repeatedly extract random fragments matched by length to the target sequences, and calculate raw scores for these fragments. The proportion of times that the raw score of a fragment set exceeds or equals the raw score of the target set, e.g. 0.02, is called a P-value. The P-value indicates the probability that the motif's presence in the target set can be explained just by chance. For each motif, a separate P-value is calculated for each background file.

The second way of determining statistical significance is to repeatedly shuffle the letters within each target sequence, and use these shuffled sequence sets as controls. P-values are calculated as above. The third way is to create random sequences with the same dinucleotide compositions as each target sequence. The fourth way is to shuffle the motif matrices, and obtain control raw scores by comparing the shuffled motifs to the target sequences. When shuffling a motif, the counts of A, C, G and T within each position are not shuffled, but the positions are shuffled among one another.

Advice

In our experience to date, the use of background sequence sets works best. However, it is necessary to choose the background sets carefully: they should ideally come from the same taxonomic group as the target sequences, and have similar repetitive element and GC content. We like to cover our bases by using multiple background sets, e.g. for human target sequences, we might use a human chromosome, a set of human CpG islands, and a set of human gene upstream regions as backgrounds. The methods that randomize nucleotides and dinucleotides suffer from predicting motifs that lie in Alus and other common repetitive elements to be significant. You should avoid including orthologous sequences from closely related species, e.g. human and mouse, as that will artefactually boost the significance of motifs in these sequences.

Output

Clover prints details for statistically significant motifs (all P-values <= some threshold, by default 0.01), and then finds instances of these motifs in the sequences. Motif instances are scored using the standard log likelihood ratio method:
score = log[ prob(sequence|motif) / prob(sequence|random) ]

Details are printed for motif instances with score >= some threshold, by default 6.

Options

In addition to the two required inputs, there several options for modifying Clover's behavior:
-h Help: print documentation.
-r Number of randomized/control raw scores to calculate for comparison with each target raw score.
-t P-value threshold: only print results for motifs whose P-values don't exceed this amount.
-u Score threshold for printing locations of significant motifs. This parameter doesn't affect raw score and P-value calculations, just which motif instances get printed.
-n Perform sequence (nucleotide) shuffles.
-d Perform dinucleotide randomizations.
-m Perform motif shuffles.
-l Mask (convert to 'n') any lowercase letters in the target sequences (and background sequences, if any). Lowercase letters are often used to indicate repetitive elements.
-v Verbose: print per-sequence scores for significant motifs. When calculating a motif's raw score, preliminary scores are first obtained for the motif compared to each sequence, and these are then combined to form the overall raw score. The -v option causes these preliminary scores to be displayed.
-p Pseudocount to add to each entry of the motif matrices. Pseudocounts are a widely used technique, with a theoretical underpinning in Bayesian statistics, for estimating underlying frequencies from a limited number of counts. If your matrices contain probabilities rather than counts, you should probably set the pseudocount to zero.
-s Seed for the random number generator (default = 1).
Example usage:

clover -t 0.05 mymotifs myseqs.fa background1.fa background2.fa

Good luck finding those motifs!

Return to Zlab Gene Regulation Hub

Suggestions to: Martin Frith
Last modified: Sunday, 20-Feb-2005 21:58:05 EST

Retrieved from "http://cagt.bu.edu/page/Clover_about"
Views
Protein Engineering