PLoS Computational Biology (2006): Transcription Factor Map Alignment of Promoter Regions

PLoS Computational Biology (2006): Transcription Factor Map Alignment of Promoter RegionsPLoS Computational Biology (2006): Transcription Factor Map Alignment of Promoter Regions

Contents
  1. ABSTRACT
  2. TF-MAP ALIGNMENT ALGORITHM
  3. MAPPING FUNCTIONS
  4. THE HR-SET DATASET
  5. TF_MAP ALIGNMENT TRAINING
  6. TF-MAP ALIGNMENT OF GENOMIC REGIONS
  7. TF-MAP ALIGNMENT OF CO-EXPRESSED GENES
  8. SOFTWARE HOME PAGE
  9. REFERENCES
ABSTRACT border=0

 

We address the problem of comparing and characterizing the promoter regions of genes with similar expression patterns. This remains a challenging problem in Sequence Analysis, because often the promoter regions of co-expressed genes do not show discernible sequence conservation. In our approach, thus, we have not directly compared the nucleotide sequence of promoters; Instead, we have obtained predictions of Transcriptions Factor Binding Sites (TFBSs), annotated the predicted sites with the labels of the corresponding Binding Factors, and aligned the resulting sequences of Transcription Factor labels--to which we refer here as Transription Factor maps (or TF-maps). To obtain the global pairwise alignment of TF-maps, we have adapted an algorithm initially developed to align restriction enzyme maps. We have optimazed the parameters of the algorithm in an small, but well curated, collection of human-mouse orthologous gene pairs. Results in this data set indicate that TF-map alignments are able to uncover conserved regulatory elements, which can not be detected by the typical sequence alignments. Exhaustive tests using the collection of human co-expressded gens in the CISRED database, further demonstrate that TF-map alignments of promoter regions are able to better discriminate co-regulated from non co-regulated genes than nucleotide sequence alignments.
 

TF-MAP ALIGNMENT ALGORITHM border=0

 

In our approach, we translate the nucleotide sequence of a promoter region S = s1 s2 ... sk into a sequence of quads A = a1, ... an where each quad ai = (aif, aip1,aip2,ais) denotes the match with score ais of a binding site for the TF aif ocurring between the position aip1 and the position aip2 over the sequence S.

The optimal alignment between two given maps A and B is the one scoring the maximum among all possible aligments. To obtain such an alignment efficiently, we have implemented an algorithm reminiscent of that proposed by Waterman et al. (1984) to align and compare enzyme restriction maps. This algorithm was developed to find the distance between two homologous restriction maps in terms of minimum weighted sum of genetic events necessary to convert one restriction map into another, where the genetic events are the appearance/disappearance of restriction sites and changes in the number of bases between restriction sites.

Here to align TF-maps A and B, we adapted such recursion to optimize similarity instead. In addition, we include a term (?) into the scoring function to weigth the scores of the matches to the TFBSs. We also discard those matches that could introduce overlap in the alignment. Thus, the maximum similarity between TF-maps A = a1, ... ai and B = b1, ... bj where the site aif is equal to the site bjf can be computed as:
 

 

Figure 1. The recurrence for computing the optimal TF-map alignment.

That is, the score of the alignment increases with the score of the matches of the aligned elements (?), and decreases with the number of unaligned elements (?), and with the difference in the distance between adjacent aligned elements (?).

The cost of the naive approach that evaluates all of the cells of the matrix when searching the best previous match for joining to the current one is O(n4). However, we propose to take advantage of the extreme sparsity of the matrix S when aligning TF-maps. In the algorithm below, we substitute the two internal nested loops by a list to register the coordinates of the match cells in the sparse matrix:
 

 

Figure 2. The enhanced algorithm.

The expected cost of this new approach can approximated to the cost of the naive algorithm divided by the length of the alphabet employed in the mapping (typically, hundreds of elements). The empirical results obtained during the program training confirmed such analysis. In average, on the order of 200 millions of elements were consulted by the naive algorithm during the optimization. In contrast, the enhanced algorithm only needed to access nearly two millions of elements to compute the same set of alignments.
 

width=300

 

Figure 3. Number of consulted elements (in millions) from the matrix S on the meta-alignment of 40 human-mouse promoter pairs. In red, the naive algorithm; in orange, enhanced algorithm with a list of computed elements; in green, enhanced algorithm with an ordered list of computed elements (click over to enlarge).

MAPPING FUNCTIONS border=0

 

In the work described in this paper, we obtain the translation from a sequence of nucleotides into a map by running on the sequence a collection of position weight matrices (PWMs) representing binding motifs for TFs. For each match over a given threshold, we register the positions, the score, and the label of the TF associated to the PWM. We will refer to the resulting map as a Transcription Factor Map (TF-map).

Four different collections of PWMs have been used in this mapping (likelihood form):
- JASPAR 1.0: 111 matrices [JASPAR database]
- PROMO 2.0: 316 matrices [PROMO database]
- TRANSFAC 6.4: 442 matrices [TRANSFAC database]
- JASPAR 1.0 TOP-50: 50 matrices

This is an example of such matrices:
 

TBP
1    61   145   152    31
2    16    46    18   309
3   352     0     2    35
4     3    10     2   374
5   354     0     5    30
6   268     0     0   121
7   360     3    10     6
8   222     2    44   121
9   155    44   157    33
10   56   135   150    48
11   83   147   128    31
12   82   127   128    52
13   82   118   128    61
14   68   107   139    75
15   77   101   140    71
        
TBP
1    -0.466    0.399    0.447   -1.143
2    -1.805   -0.749   -1.687    1.156
3     1.286   -5.270   -3.884   -1.022
4    -3.479   -2.275   -3.884    1.347
5     1.292   -5.270   -2.968   -1.176
6     1.014   -5.270   -5.270    0.219
7     1.335   -3.453   -2.249   -2.759
8     0.825   -3.884   -0.793    0.219
9     0.466   -0.793    0.479   -1.081
10   -0.552    0.328    0.433   -0.706
11   -0.158    0.413    0.275   -1.143
12   -0.171    0.267    0.275   -0.626
13   -0.171    0.193    0.275   -0.466
14   -0.358    0.096    0.357   -0.260
15   -0.233    0.038    0.364   -0.315

 

Figure 4. On the left, a frequency matrix (M). On the right, a likelihood matrix (LM)

 

Each row in the frequency matrix M corresponds to the observed distribution of nucleotides in this position of the motif after an aligment of real sites was done. Thus, the element M(x,i) in the matrix is the number of cases in which the nucleotide x was observed at position i. The probability or score to observe such fact is obtained with P(x,i) = M(x,i) / M(A,i) + M(C,i) + M(G,i) + M(T,i).

Each element LM(x,i) in the likelihood matrix LM corresponds to the logarithm of the ratio between the probability of observing such a nucleotide in this position in a real site versus the probability of observing it in a false site. False site distribution is modeled with a uniform random distribution (0.25,0.25,0.25,0.25). The maximum score MAX_SCORE of a matrix is the sum of the highest score at each row. The scoring method for a segment S=s1s2...sn with a matrix LM is then:

 

Figure 5. Computing the score of a mapped binding site in a sequence

 

 

MATSCAN
The mapping functions described here are implemented in the program matscan v1.0. MATSCAN searches for occurrences of binding sites or any type of genomic signal that can be defined using weight matrix models. The predictions are displayed in GFF format, being graphically represented with the program gff2ps. The source code of the program is publicly available at the home page of the software implementing the TF-map alignment: [MatScan source code] [Examples]

The program matscan v 1.0 was written by Enrique Blanco and Roderic Guigo.

If you use this program, please cite:
E. Blanco, X. Messeguer, T.F. Smith and R. Guigo.
Transcription Factor Map Alignment of Promoter Regions.
PLoS Computational Biology 2(5): e49 (2006).
 

THE HR-SET DATASET border=0

 

We have gathered and manually curated a collection of 278 TFBSs (139 + 139 orthologous sites) that had been experimentally tested in 40 orthologous human and rodent genes (ABS, Blanco et al., 2006). Because most (214 out of 278) of the annotated TFBSs are located in the 200 nucleotides inmediately upstream the TSS, we restricted to this region in our training and evaluation analysis, and considered only those cases for which the same pair of TFBSs had been annotated in this region for both species. This resulted in a collection of 202 sites (101 + 101) from 36 genes, to which we refer here as the HR-set.

40 human-mouse gene promoter pairs:
-The 40x2 promoters of the collection of 278 TFBSs (500 bps upstream the TSS)
-The collection of 278 TFBSs

The HR-set (36 human-mouse gene pairs):
-The 36x2 promoters of the collection of the HR-set (200 bps upstream the TSS)
-The HR-set (202 TFBSs, 101 + 101)

MAPPING on the 40x40 promoter sequences:
-JASPAR predictions
-PROMO predictions
-TRANSFAC predictions
-JASPAR TOP-50 predictions

ANNOTATIONS SUPPORTED BY ANY PREDICTION:
-186 annotations supported by JASPAR predictions
-188 annotations supported by PROMO predictions
-188 annotations supported by TRANSFAC predictions
-100 annotations supported by JASPAR TOP-50 predictions
 

width=250

 

Figure 6. Positional distribution of the 278 TFBSs along the 500 bps of the 40x2 promoters (click over to enlarge).

TF-MAP ALIGNMENT TRAINING border=0

 

The optimal alignment between two TF-maps is obviously dependendant on the ?, ? and ? parameters. We have estimated the optimal parameters in the HR-set for the JASPAR 1.0, PROMO 2.0 and TRANSFAC 6.3 collections.

OPTIMAL CONFIGURATIONS FOR EACH MAPPING FUNCTION:
-Best configuration using JASPAR
-Best configuration using PROMO
-Best configuration using TRANSFAC
-Best configuration using JASPAR TOP-50

For each configuration, the correctly detected pair of sites on each pair of sequences is denoted with OK. The length and the score of the meta-alignment are also provided. At the end, the accuracy measures at site level, site+label level and nucleotide level are displayed (see main text).

NEGATIVE CONTROL BY SHUFFLING THE HR-set PROMOTERS:
-Original orthology associations (36 genes)
-Associations after shuffling (36 genes)
-Score and length of the 36 shuffled TF-map alignments using JASPAR
-Score and length of the 36 shuffled BLASTN sequence alignments
 

TF-MAP ALIGNMENT OF GENOMIC REGIONS border=0

 

To assess to what extend good TF-map alignments are a simply reflection of underlying sequence conservation, we have compared the meta-alignments obtained using JASPAR-TOP 50, in the 200 bp promoter region of the 36 gene pairs from the HR-set, with the meta-alignments obtained in 200 bp fragments from intergenic (2000bp upstream of the TSS), 5'UTR (downstream of the TSS), coding (downstream of the translation start site and considering only coding DNA), intronic (downstream of the first intron junction), and downstream (downstream of the transcription termination site). We have computed the average score of the map alignments in each of the genomic regions and have identified for each homologous pair, the genome regions in which the alignment produces the highest score. We have performed the same exercise using global pairwise sequence alignments (obtained with CLUSTALW).

In addition, we have performed the same comparison on the human-chicken orthologous pairs of the HR-set. According to the UCSC genome browser, we found the chicken ortholog in REFSEQ for 25 human genes of the HR-set. Then, we compared again the meta-alignments obtained using JASPAR-TOP 50, in the 200 bp promoter region of the 25 gene pairs from the HC-set, with the meta-alignments obtained in 200 bp fragments from intergenic (2000bp upstream of the TSS), 5'UTR (downstream of the TSS), coding (downstream of the translation start site and considering only coding DNA), intronic (downstream of the first intron junction), and downstream (downstream of the transcription termination site). We also have performed the same exercise using global pairwise sequence alignments (obtained with CLUSTALW).

HR-set:
- The HR-set (36 gene pairs)
- Samples of the human and rodent genomic regions
- TF-map alignment results using JASPAR-TOP 50
- Sequence alignment results using CLUSTALW

HC-set:
- The HC-set (25 gene pairs)
- Samples of the chicken genomic regions
- TF-map alignment results using JASPAR-TOP 50
- Sequence alignment results using CLUSTALW
 

TF-MAP ALIGNMENT OF CO-EXPRESSED GENES border=0

 

We have performed an exhaustive evaluation of the power of TF-map alignments to characterize promoter regions of co-regulated genes in absence of sequence similarity. Towards such a goal we have used the set of high-confidence coexpressed gene pairs (Griffith et al,2005), obtained from cDNA microarray hybridization, SAGE and other experiments, as well as Gene Ontology analysis included in the CISRED database. Each group is always identified by a representative gene and the set of co-regulated genes with it.
 

width=300 width=300

 

Figure 7. Gumbel distribution of the TF-map alignment scores: (on the left) distribution of scores in the alignments involving co-regulated genes; (on the right) distribution of scores in the alignments involving non co-regulated genes (click over to enlarge).

 

DATA:
- The 5333 CISRED genes with available promoter
- The promoter sequence of the available 5333 CISRED genes (500 bps)
- The (-100:100) sequence around the TSS annotated by BIOMART of the CISRED genes (200 bps)
- CpG+ and CpG- classification of the CISRED genes: [gene,GC content,CpG score, class]
 

 

Figure 8. CpG islands: G+C content and CpG score

 

TF-MAP alignments:
- TF-map alignment results between co-regulated genes (21288x2 alignments, best HSP score reported)
- TF-map alignment results between non co-regulated genes (10300320x2 alignments, best HSP score reported)
- 1240 TF-map alignments between co-regulated genes that are in TOP-445,371 best alignments
- Results of evaluating the significance of TF-map alignments (Wilcoxon test p-value)
- 97 groups with p-value < 0.01 and 23 groups with p-value < 0.001

Discarding the presence of recent duplications in the 97 genes set:
- The 97 gene groups with p-value < 0.01
- 97 x 97 BLASTN promoter alignments: best HSP for each pair (avg. score: 32.14, avg. length: 59.91)
- 97 x 97 NEEDLE (EMBOSS software) promoter alignments (avg. %similarity: 36.97)

BLASTN alignments:
- BLASTN alignments between co-regulated genes (981 alignments, best HSP score reported)
- BLASTN alignments between non co-regulated genes (444390 alignments, best HSP score reported)
- Results of evaluating the significance of BLASTN alignments (653 groups, Wilcoxon test p-value)
* 11 groups with p-value < 0.01 and 1 group with p-value < 0.001

RANKING - Coreg alignments in the TOP-445,371:
- Intersection between the 981 BLASTN alignments and the 1240 TF-map alignments
- Intersection between the 1240 TF-map alignments and the 1072 pairs detected by TF conservation coefficient

TTR promoter reconstruction:
- The 83 TF-map alignments between the TTR gene and a co-regulated gene
- The reconstruction of the TTR promoter using the 83 TF-map alignments
 

SOFTWARE HOME PAGE border=0

 

The home page of the software implementing the TF-map alignment is at:
TF-MAP ALIGNMENT HOME PAGE

Here you can find:
- The source code of the program
- Benchmarking test
- Examples of the output

The web server that performs the mapping and the TF-map alignment is at:
TF-MAP ALIGNMENT WEB SERVER

The available functions now are:
- Mapping applications to translate a promoter sequence into a TF-map with different collections and thresholds
- Meta-alignment of two TF-maps allowing the user to configure the parameters
- Dummy example

REFERENCES border=0

 

  • E. Blanco, X. Messeguer, T.F. Smith and R. Guigó

    Transcription Factor Map Alignment of Promoter Regions

    PLoS Computational Biology 2(5): e49 (2006).
     
  •  

  •  
  •  
  •  
  •  
  •  
  •  
  • E. Blanco, D. Farre, M. Alba, X. Messeguer and R. Guigó

    ABS: a database of Annotated regulatory Binding Sites from orthologous promoters.

    Nucleic Acids Research 34:D63-D67 (2006).
     
  • E. Blanco, X. Messeguer and R. Guigó

    Novel computational methods to characterize regulatory regions.

    CSHL - Systems Biology (Poster), New York, USA (2004)
     
  • E. Blanco, X. Messeguer and R. Guigó

    Alignment of Promoter Regions by Mapping Nucleotide Sequences into Arrays of Transcription Factor Binding Motifs.

    VIIth RECOMB (Poster), Berlin, Germany (2003)
     
  • M.S. Waterman, T. F. Smith and H. L. Katcher

    Algorithms for restriction map comparisons,

    Nucleic Acids Research, 12:237-242 (1984).
     
  • O.L. Griffith, E.D. Pleasance, D.L. Fulton, M. Oveisi, M. Ester, A.S. Siddiqui and S.J. Jones.

    Assessment and integration of publicly available SAGE, cDNA microarray, and oligonucleotide microarray expression data for global coexpression analyses.,

    Genomics,86(4):476-488 (2005).
Group: