SinicView offers a series of manipulative and navigational controls, such as zooming, shifting, and gap/annotation toggling. As shown in Figure 1, SinicView displays the alignment results obtained by three different MSA methods. The input sequences contain the orthologous regions around the Stem Cell Leukemia (SCL) gene in five vertebrate species: human, mouse, chicken, pufferfish and zebrafish. The buttons and text-field boxes of manipulative functions are located on top of the frame. Users can manually input numerical values or click on the highlighted colored region in the Global View section that specifies the zooming or shifting factors in a drag-and-drop fashion. When the highlighted region is clicked and dragged, the equivalent of a shift action will be performed and the display region can be resized by adjusting the edge of the highlighted area.

SinicView can display more than one alignment result obtained by different alignment programs (either pairwise or multiple ones.) The assorted mixed-color span under the Global View panel shows among the alignment tools used the preferred aligner, which generates comparatively better results on the spot. Each of the aligners is denoted by a pre-defined color with the "performance color" label right next to the name of the tool.

In the Detailed View section, the Percent Identity Plot (PIP) panels show, from top to bottom, the similarity curves of the alignment results obtained by different programs, along with the names of the alignment tools. In the Information View section, the Gap & Annotation panels (in pink and gray) display the information of annotations provided by users, and gaps of aligned sequences. The information and similarity ratios can also be displayed as the current scan-line (i.e. cursor) moves. The boxes in maroon denote the annotation area and the horizontal line represents the original sequences interleaved with inserted gaps (light gray areas.) The gap display can be toggled on or off via the checkbox on the right.

Because different alignment results are usually of different lengths, it is not plausible to compare these results base-pair by base-pair. In SinicView, therefore, we let users select one of input sequences as a reference and then calculate the sum-of-pair scores of each base pair in the reference within a fixed window. For example, each alignment result in the PIP panels at the scan-line position corresponds to human sequence, selected as the reference in Figure 1. When the user selects different sequences as the reference, SinicView can demonstrate the variations between the PIP curves of the alignment results.

The functionality under the "Tools" menu, called "Comparison Charts", offers two types of charts for quick-and-easy evaluation of the alignment quality. The stacked bar chart, in Figure 2, illustrates the distribution of the identical rates with the threshold over 40%. The pie chart, on the other hand, displays the distribution of the identical rates from 0 to 100 percent based upon a selected alignment program. The statistics on which these charts are based can also be displayed in a tabulated text form.

SinicView also provides a plain-text view of the alignment results in the Detailed View section when the sliding window size is less than 100 aligned base pairs. As shown in Figure 3, the plain-text alignment results replace the percent identity curves and the fully identical bases in a column are labeled in red blocks. Thus, users can check the correctness of detailed alignment results base pair by base pair.

The applet version can be accessed via any JRE (Java Runtime Environment)-enabled browsers with Internet connection, thus making the installation and choosing the right platform hassle-free. However, the ease of running SinicView on-the-go cannot accommodate the bandwidth requirement in case of huge amount of sequence data involved. Hence, we have also implemented a standalone application of SinicView, which is wrapped in JRE, for off-line use.

The execution procedure of the standalone SinicView is quite straightforward. Upon launch, the user will be prompted three options. The first two are to read user's Phylogenetic Tree files, an option, and MSA results from the local disk.

The Stem Cell Leukemia (SCL) gene plays a critical role in normal processes that, when disrupted, can result in leukemia. The SCL gene, also known as tal-1, encodes a basic helix-loop-helix transcription factor that is pivotal for the normal development of all hematopoietic lineages, and is highly conserved between mammals and zebrafish 5152. Previous analyses of the SCL genes in five vertebrate genomes, including human, mouse, chicken, pufferfish, and zebrafish, have revealed that the SCL promoter/enhancer motifs are conserved in all five species 51. The alignment and visualization tools used in their analyses included BLAST 53, PipMaker 45, and DiAlign 54. Shah et al. (2004) realigned these gene regions in five species by a pairwise alignment tool, LAGAN 20, and demonstrated the alignment result by Phylo-VISTA 44. In this paper, we also downloaded these sequences and realigned them by the multiple alignment tools: ClustalW, MAVID and MLAGAN. The lengths of the human, mouse, chicken, pufferfish, and zebrafish sequences are approximately 100 kb, 65 kb, 67 kb, 22 kb, and 8 kb, respectively.

Figure 4(a) shows the global view of the results obtained by three alignment tools using the human sequence as the reference. Generally speaking, the highest conserved region located at 30 k bp of human sequence is all well aligned by these three tools. But the highest identical rates of the alignment by ClustalW are lower than those by either MLAGAN or MAVID. Moreover, the total quantity of the result obtained by MLAGAN is better than those by both ClustalW and MAVID while the quantity of the result obtained by ClustalW is better than those by the others, as shown in Figure 4(b). Interestingly, when we selected the zebrafish sequence as the reference, the result obtained by ClustalW shows the highest conserved region located at around 27.5 k bp whereas those by both MAVID and MLAGAN show it at around 45.89 k bp, as shown in Figure 4(c). The comparison reveals that the region at around 27.5 k bp in the zebrafish sequence will be assumed the homologous region by ClustalW. But according to MAVID and MLAGAN, the homologous regions are located at around 45.89 k bp rather than at 27.5 k bp. This ambiguous result may be caused by segmental duplication in the sequences and by difference in alignment strategy. In this case, more advanced or further inspections should be performed to either check the detailed alignment results in both regions or realign these sequences by using other pairwise or local alignment tools.

The cystic fibrosis transmembrane conductance regulator (CFTR) gene is responsible for the cystic fibrosis disorder that spans approximately 190 k bp of genomic DNA and consists of 27 exons 55. The greater CFTR region is defined as a genomic segment of about 1.8 M bp on human chromosome 7q31.3 containing the CFTR gene and nine other genes, including TES1, CAV1, CAV2, MET, CAPZA2, ST7, WNT2, GASZ, and CORTBP2 12. The comparative analysis of this region in 13 vertebrate species has been reported in Thomas et al., 2003 12 in which the alignment tool used was BlastZ on PipMaker Web server 45. In this paper, we downloaded the sequences of four mammalian species, including human, baboon, dog, and mouse, from the NIH Intramural Sequencing Center (NISC) Website 56. However, the original sequences had been updated in other genome browsers. Thus, we eventually downloaded the last versions of these sequences from the UCSC Genome Browser. The lengths of these sequences are from 1.0 M bp to 1.5 M bp. We realigned these sequences by MLAGAN, MAVID, and TBA (kernel: MULTIZ) 24 and the total number of bases of the final alignment results, including gaps, are approximately 12 M bp, 11 M bp, and 7.5 M bp, respectively.

Figures 5(a) and 5(b) show the global PIP curves and their detailed views of three alignment results, respectively. In general, most of high identity regions are well and consistently aligned by these three programs. But those not as high identities are not reported by TBA because the kernel of this program, MULTIZ, is based on the local alignment results by BlastZ. As shown in Figure 5(c), the stacked-bar charts show the quality and the quantity of these alignment results where the average identical rates for TBA are somewhat better than those for MLAGAN and MAVID although the total number of aligned conserved regions for MLAGAN is larger than those for the others.

For comparisons of these alignments from a functional viewpoint, we downloaded the annotation of the human sequence, including exons and repeats, from the Ensembl Genome Browser 35. The detailed comparisons of the alignment results by different aligners demonstrated that the alignments of noncoding regions are often inconsistent. But for the coding regions, the alignment results by different aligners seem consistent and well-aligned.

Figures 6(a)–(b) show the detailed alignment results at four different intervals. In Figure 6(a), we find that some conserved regions are not aligned by TBA but identified by MLAGAN and MAVID. This region is annotated by repeats and implies that some repetitive elements were inserted into these sequences of their common ancestor. However, this conserved insertion event could not be observed by using TBA. Although the kernel of TBA, MULTIZ, is known not to align regions with repetitive elements, we still find that some other regions with repetitive elements are aligned by this program, as shown in Figure 6(b).

Generally speaking, the regions aligned by TBA usually have higher identical rates than by others. As the frames shown in red in Figures 6(c) and 6(d), the alignment of these regions by TBA seems superior to those by others. However, the kernel of TBA, MULTIZ, usually neglects to align the regions with low conservations. Thus, some lowly conserved regions may not be aligned by TBA.

Since each alignment tool has its own advantage and reveals different alignment results, we therefore wonder whether a better alignment result can be generated by hybridization of these alignment tools.

SinicView is implemented totally in Java. Theoretically, it should be portable across different operating systems (OSs) and platforms. To demonstrate interoperability on real cases, we tested the applet and application versions of SinicView on different platforms and OSs. As shown in Table 1, both versions of SinicView seem to perform well. Thus, users can use either the applet version or the standalone application of SinicView, according to their requirements.

Besides, we also tested the loading performance of SinicView. Because the performance of an applet on the Web is strongly dependent on the network bandwidth and traffic, the estimation of loading time may not be a fair comparison. Thus, in this part we only estimated the loading performance of the standalone application of SinicView.

In general, the loading performance of a Java application is dependent on the memory heap size. The default values of the initial heap size and the maximum size of a Java Virtual Machine (java_1.4.2 version or higher) are 4 M (mega) bytes and 64 M bytes, respectively. These values can be adjusted by the following command in the terminal mode:

java -Xms64m -Xmx128m -jar SinicView.jar,

where the parameters Xms64m and Xmx128m represent that the initial heap size is 64 M bytes and the maximum size is 128 M bytes, respectively. Thus, we used different input data sizes, initial heap sizes, and the maximum sizes to estimate the loading time of SinicView. As shown in Table 2, using the default maximum heap size, 64 M bytes, the standalone SinicView can handle up to approximately 11 M bytes alignment data. If the maximum size is set up to 256 M bytes, the loading ability of input data size could be over several dozens of mega bytes. Moreover, Table 2 shows that the maximum data size is dependent on the maximum heap size and the loading times are linearly dependent on the sizes of input data. All performance test results were benchmarked on a 3 GHz Pentium4 PC with 1 GB RAM.

The eukaryotic genome is usually characterized by the presence of repetitive DNA consisting of nucleotide sequences of various lengths and compositions that occur from a few times to thousands of times in the genome either in tandem or in a dispersed fashion57. The repetitive fractions can be classified into two types of repeated families: localized and dispersed 5758. Localized repetitive sequences usually occur as tandem arrays and they are called tandem repetitive DNA. Dispersed repetitive sequences are dispersed throughout the genome. In addition, there are moderately repetitive sequences, which are usually transposable elements or processed pseudogenes and are usually dispersed over the genome. Alu is the largest family of interspersed mobile elements (~300 bp) and propagated to more than one million copies in primate genomes. This type of repeat has been inserted into these genomes within the last 65 million year period 58. Because this type of repetitive elements only appears in the primate genomes, when we align homologous sequences of primate and non-primate genomic sequences, these Alu inserted regions should not be aligned. However, other interspersed elements may possibly have been inserted into the ancestral sequence of mammalians. The regions of these repeats may be able to align together between the sequences of different mammalians, as shown in Example 2. However, these regions in the alignment results by different aligners are inconsistent. Since these repetitive elements in sequences could be detected by RepeatMasker 59, the poorly aligned regions may have to be checked whether they belong to repetitive elements.

As the comparison results using SinicView show, the alignments of sequences using different MSA tools are inconsistent. We begin to wonder whether the computational results obtained by different tools may in fact lead to different findings. For identification of alignment correlation, a need for additional checks of alignment validity by using different tools and scoring systems has been recognized in the literature 60. Thus, a cross comparison approach along with visualization could provide an efficient and easy way for general users to verify and validate the alignment results as to whether the aligned regions are reasonable and whether those poorly aligned regions are indeed non-homologous.

Except evaluation of the alignment quality by comparison charts in SinicView, how to decide on a good alignment with biological meanings may need much more experiences and knowledge. Sometimes, this judgment depends also on what kind of the biological problems users want to study. Here, we suggest some general rules for users to judge the alignments by biological meanings.

In the coding regions, a triplet of adjacent nucleotides constitutes a codon. Usually, the first two nucleotides are identical between the two sequences and allow the third one to be either identical or different. Thus, when the partial alignment results reveal the two-out-of-three regularity for each triplet, it may imply that the aligned regions are potential coding regions. This alignment result should be more biologically meaningful than those without the two-out-of-three regularity.

From molecular evolutionary viewpoint, nature prefers inserting or deleting considerable consecutive nucleotides together to interspersed individual nucleotides 57. Thus, an alignment with consecutive gaps would be better than those with interspersed gaps.

If one of the alignment sequences has been annotated, the information is definitely useful for users to judge the alignment results by different aligners.

It is not easy to promote newly developed tools because users usually cannot directly compare the new tools with the traditional ones. With SinicView, users can compare the alignment results obtained by different tools and select an appropriate one for further analysis. Thus, if the new tool can align more regions than those by the old ones and can also indicate their statistical significances, it will be welcomed and better received by the community. We would like to make SinicView available to the community of computational biologists. In addition to helping the user find a most appropriate alignment tool to use, SinicView may also be used to check whether previously obtained alignment results by different tools are worth a re-investigation, and see if this revisit of alignment results would lead to different conclusions.

The capability of fine-tuning parameters relevant to the alignment process will be made available in a user-friendly interface. Furthermore, the ability to allow plug-ins of more alignment programs, in addition to the currently pre-selected ones, such as ClustalW, MAVID, MLAGAN, and GS-Aligner, will inevitably broaden the usage of SinicView. The issue of the compatibility of the input and output formats for each alignment tool also needs to be resolved. For example, both MAVID and MLAGAN require the phylogenetic tree data as input, but ClustalW does not. The ordering of the outputs of these aforementioned tools is usually switched without notice. Thus, to be able to work under a unified comparison framework requires further processing of these outputs. Besides, identifying a standard-bearer mechanism is still a challenge in entrusting existing alignment programs. So far, we have used the "sum-of-pairs" method to define the "identical rate" in each alignment result. In the future, we may provide other criteria for users to use to measure their alignment results, in addition to what have been already provided in SinicView.