Kornelie Frech

MatInspector and beyond: promoter analysis based on transcription factor binding sites

MOTIVATION Promoter analysis is an essential step on the way to identify regulatory networks. A prerequisite for successful promoter analysis is the prediction of potential transcription factor binding sites (TFBS) with reasonable accuracy. The next steps in promoter analysis can be tackled only with reliable predictions, e.g. finding phylogenetically conserved patterns or identifying higher order combinations of sites in promoters of co-regulated genes. RESULTS We present a new version of the program MatInspector that identifies TFBS in nucleotide sequences using a large library of weight matrices. By introducing a matrix family concept, optimized thresholds, and comparative analysis, the enhanced program produces concise results avoiding redundant and false-positive matches. We describe a number of programs based on MatInspector allowing in-depth promoter analysis (DiAlignTF, FrameWorker) and targeted design of regulatory sequences (SequenceShaper).

DIALIGN: finding local similarities by multiple sequence alignment.

MOTIVATION DIALIGN is a new method for pairwise as well as multiple alignment of nucleic acid and protein sequences. While standard alignment programs rely on comparing single residues and imposing gap penalties, DIALIGN constructs alignments by comparing whole segments of the sequences. No gap penalty is employed. This point of view is especially adequate if sequences are not globally related, but share only local similarities, as is the case in genomic DNA sequences and in many protein families. RESULTS Using four different data sets, we show that DIALIGN is able correctly to align conserved motifs in protein sequences. Alignments produced by DIALIGN are compared systematically to the results of five other alignment programs. AVAILABILITY DIALIGN is available to the scientific community free of charge for non-commercial use. Executables for various UNIX platforms including LINUX can be downloaded at http://www.gsf.de/biodv/dialign.html CONTACT werner, morgenstern@gsf.de

Functional promoter modules can be detected by formal models independent of overall nucleotide sequence similarity.

MOTIVATION Gene regulation often depends on functional modules which feature a detectable internal organization. Overall sequence similarity of these modules is often insufficient for detection by general search methods like FASTA or even Gapped BLAST. However, it is of interest to evaluate whether modules, often known from experimental analysis of single sequences, are present in other regulatory sequences. RESULTS We developed a new method (FastM) which combines a search algorithm for individual transcription factor binding sites (MatInspector) with a distance correlation function. FastM allows fast definition of a model of correlated binding sites derived from as little as a single promoter or enhancer. ModelInspector results are suitable for evaluation of the significance of the model. We used FastM to define a model for the experimentally verified NFkappaB/IRF1 regulatory module from the major histocompatibility complex (MHC) class I HLA-B gene promoter. Analysis of a test set of sequences as well as database searches with this model showed excellent correlation of the model with the biological function of the module. These results could not be obtained by searches using FASTA or Gapped BLAST, which are based on sequence similarity. We were also able to demonstrate association of a hypothetical GRE-GRE module with viral sequences based on analysis of several GenBank sections with this module. AVAILABILITY The WWW version of FastM is accessible at: http://www.gsf.de/cgi-bin/fastm. pl and http://genomatix.gsf.de/cgi-bin/fastm2/fastm.pl

Identification of functional elements in unaligned nucleic acid sequences by a novel tuple search algorithm

We present an algorithm to identify potential functional elements like protein binding sites in DNA sequences, solely from nucleotide sequence data. Prerequisites are a set of at least seven not closely related sequences with a common biological function which is correlated to one or more unknown sequence elements present in most but not necessarily all of the sequences. The algorithm is based on a search for n-tuples which occur at least in a minimum percentage of the sequences with no or one mismatch, which may be at any position of the tuple. In contrast to functional tuples, random tuples show no preferred pattern of mismatch locations within the tuple nor is the conservation extended beyond the tuple. Both features of functional tuples are used to eliminate random tuples. Selection is carried out by maximization of the information content first for the n-tuple, then for a region containing the tuple and finally for the complete binding site. Further matches are found in an additional selection step, using the ConsInd method previously described. The algorithm is capable of identifying and delimiting elements (e.g. protein binding sites) represented by single short cores (e.g. TATA box) in sets of unaligned sequences of about 500 nucleotides using no information other than the nucleotide sequences. Furthermore, we show its ability to identify multiple elements in a set of complete LTR sequences (more than 600 nucleotides per sequence).

Software for the analysis of DNA sequence elements of transcription

The detection of transcription control elements in DNA sequences became both more important and more complicated by the completion of the first full genome sequencing projects. Rapid evaluation of potential regulatory elements in large amounts of sequence data requires specific methods preferably available as user-friendly computer programs. However, many more algorithms and methods have been published than programs are available, creating problems for scientists who try to select an appropriate method for their needs from the literature. The Internet provides a worldwide and relatively easy access to computer software if the user knows where to look. One of the major problems remaining is how to find the appropriate software. We have compiled a guide detailing where software is available and what is to be expected in terms of interface and data compatibility with other programs. We also show results obtained with each program for several examples. The summarized features of each program should allow scientists to select quickly the method of their choice and inform them where to download the software.

LitInspector: literature and signal transduction pathway mining in PubMed abstracts

LitInspector is a literature search tool providing gene and signal transduction pathway mining within NCBIs PubMed database. The automatic gene recognition and color coding increases the readability of abstracts and significantly speeds up literature research. A main challenge in gene recognition is the resolution of homonyms and rejection of identical abbreviations used in a ‘non-gene’ context. LitInspector uses automatically generated and manually refined filtering lists for this purpose. The quality of the LitInspector results was assessed with a published dataset of 181 PubMed sentences. LitInspector achieved a precision of 96.8%, a recall of 86.6% and an F-measure of 91.4%. To further demonstrate the homonym resolution qualities, LitInspector was compared to three other literature search tools using some challenging examples. The homonym MIZ-1 (gene IDs 7709 and 9063) was correctly resolved in 87% of the abstracts by LitInspector, whereas the other tools achieved recognition rates between 35% and 67%. The LitInspector signal transduction pathway mining is based on a manually curated database of pathway names (e.g. wingless type), pathway components (e.g. WNT1, FZD1), and general pathway keywords (e.g. signaling cascade). The performance was checked for 10 randomly selected genes. Eighty-two per cent of the 38 predicted pathway associations were correct. LitInspector is freely available at http://www.litinspector.org/.

ConsInspector 3.0: new library and enhanced functionality.

Conslnspector (Freeh et al, 1993) is a program to scan nucleic acid sequences for matches to a pre-compiled library of transcription factor binding sites. The program carries out an extensive examination of binding site candidates; the real sequence is compared with randomly shuffled versions and sequence regions surrounding the conserved binding site are included in the analysis (default 40 bp upstream and 40 bp downstream of the highly conserved core sequence). This feature distinguishes the program from other methods available for the identification of transcription factor binding sites which are restricted to the binding sites: SIGNAL SCAN (Prestridge, 1991, 1996; Prestridge and Stormo, 1993), MATRIX SEARCH (Chen et al, 1995) and Matlnspector (Quandt et al, 1995a). Recently, we showed the quality scores (Q-scores) assigned by Conslnspector to correlate to some extent with biological functionality (Quandt et al, 1995b). Release 3.0 of Conslnspector, with enhanced performance and a considerably extended library of consensus profiles, is available now at ftp://ariane.gsf.de/pub/ or http://www.gsf.de/biodv/. The program Conslnd (Freeh et al, 1993) has been used to compile the library of consensus profiles. The library now encompasses 37 consensus profiles (Release 1.0: 12, Release 2.1: 17 consensus profiles) and is separated into four groups (Table I). The extended weight matrices were deduced from experimentally confirmed binding sequences selected from the TRANSFAC database (Wingender et al, 1996) or directly from the literature. Most consensus profiles of the original library have been improved by the inclusion of additional sequences. Consensus profiles have been compiled from a minimum of nine sequences (Table I). The analysis of DNA sequences for transcription factor binding sites with Conslnspector has improved since Release 1.0:

A New Method to Develop Highly Specific Models for Regulatory DNA-Regions

We present a new modular concept to construct organizational models for transcriptional regulatory DNA units. The method requires a training set of at least 10 sequences and a simple initial model (e.g. two characteristic transcription factor binding sites). The final model is generated by computer analysis directly from the sequences. 20 Lentivirus long terminal repeats (LTRs) and an initial model consisting of only two elements (TATA box and polyA signal) resulted in a final model of 10 elements which recognized all of the more than 100 available Lentivirus LTRs while rejecting all other known LTR types. Database searches with this Lentivirus LTR model demonstrated the very high specificity of our method.