1. Introduction

In 2001, the Human Genome Project (HGP) Consortium and Celera Genomics reported the first drafts of sequences of the human genome [1, 2]. The HGP Consortium used the hierarchical sequencing or “clone-by-clone” approach, whereas Celera Genomics used the whole genome shotgun (WGS) approach, which had been successfully used in 1995 to sequence the H. influenzae genome [3].

In the hierarchical sequencing approach, a tiling of large DNA sequences, such as bacterial artificial chromosome (BAC) or yeast artificial chromosome (YAC), are constructed for a genome, and each of the sequences is determined. The HGP Consortium used BAC as the large sequence, followed by shotgun sequencing of each BAC.

In sequencing the genome, owing to physical limitations of shotgun sequencing methods, the genome must be broken down into smaller portions, shotgun reads sized in the range of 600 bps (base-pairs) to 800 bps, and as the sequence data for each of these shotgun reads is produced, it must be connecting them with those adjacent and overlapping reads that have been previously sequenced, that is, to achieve an assembly of these smaller sequences into larger contiguous regions or “contigs.”

In most cases, the sequences of shotgun reads are obtained by sequencing both ends of a DNA fragment whose approximate size is known. Such pair information, referred to as mate-pair information, constrains the placement of the reads within an assembly. In an ideal assembly, all read pairs are placed in such a manner as to satisfy the orientation and distance constraints imposed by the pairing. Mate-pair information can be used to determine the quality of an assembly, because most types of misassemblies lead to violations of these constraints.

In contrast to hierarchical sequencing, WGS breaks a whole genome into small pieces randomly, without shearing into large DNA pieces of intermediate size. WGS is faster and cheaper than hierarchical sequencing because of the simplicity of the processing steps. The success of WGS [4, 5] has increased its usage and the size of the genome to be sequenced has increased.

Although contig assembly programs are well established, less is known about scaffold analysis. While some of its features have been implemented to sequence specific genomes [6–8], the features needed for general scaffold analysis and visualization have not been provided. Consed [9], a graphical tool for contig assembly, provides good visualization and helps to finish sequencing by connecting with Autofinish [10]; however, it does not have many features related to scaffold analysis.

It has been suggested that the contig scaffolding problem can be solved by greedy-path merging algorithm [8]. Moreover, GigAssembler can orient the contigs based on mRNA, paired plasmid ends, EST, and BAC end pairs [7].

This paper introduces a novel scaffold analysis tool, ConPath, which calculatesthe longest scaffolds. Due to the abundance of repeats in genomic DNA sequences, a purely overlap-based approach for WGS assembly is not feasible, but the use of mate-pair information is crucial. The ConPath program uses end read pairs of fixed-sized DNA libraries as mate-pairs to calculate orientations, orders, and gap sizes. It reads a Phrap [11] output file (*.out) and an ACE format file, which contain contig structures and mate-pair information.

2. Materials and Methods

2.1. Mate-Pair Information

The most important characteristic of ConPath is its ability to exploit the mate-pair information of large DNA fragments such as fosmids or cosmids, which are about 40 kbps(kilo base-pairs) in size, or BACs, which are about 100–300 kbps in size, rather than plasmids, which are about 2–10 kbps in size. Figure 1 shows an example of mate-pair end reads. A mate-pair is composed of two end reads that always face each other. Each end read, or , has an orientation relative to the contig containing it. If the direction of an end read is the same as the direction of the contig, the former has direction , otherwise, it has direction . In Figure 1, has direction because the contig and read are in the same direction, whereas has direction because the contig and read are in opposite directions. The size of the mate-pair helps to estimate the gap size between contigs and . When one contig contains one end of a mate-pair and a second contig contains the other end of the mate-pair, the two contigs are said to be linked by the mate-pair. A scaffold is a series of contigs that can be linked by mate-pairs. The connection relationship of all the contigs can be represented as a graph in which each contig is represented as a vertex. An edge is created between two contig vertices when they are linked by at least one mate-pair, and the number of linking mate-pairs between two contigs is defined as the edge weight.

Figure 1: An example of mate-pair information. Mate-pair reads are indicated as read “” and “” and the relative directions to encompassing contigs are denoted as “ (same direction)” and “ (complementary direction).”

2.2. Construction of Scaffolds

To construct scaffolds using mate-pair information, a scaffold graph can be defined as follows.

Given a set of contigs a mate-pair set and a set of reads let denote the scaffold graph using and :

When a mate-pair exists, in which contig contains and contig contains there is an edge between contigs and Edge set is expressed as

In constructing a scaffold graph, the linking level the threshold value for the edge weights, was used as a filtering value in constructing and showing scaffolds on output. When an edge has a weight value smaller than the linking level the edge is discarded from the graph.

Considering the errors that occur in base calling and contig assembly, the optimal construction of a scaffold graph is an NP-complete problem [8]. To practically solve this problem, ConPath uses a simple greedy algorithm. Whenever a new edge is added to the graph, graph is additive modified for that edge. This provides a feasible heuristic solution for a scaffold construction in linear time. Algorithm 1 shows the algorithm of ConPath to construct scaffolds.

Algorithm 1: The algorithm for scaffold construction. ConPath uses a simple greedy algorithm to obtain a feasible heuristic solution for an NP-complete problem.

2.3. Determination of the Orders and Orientations of Contigs

It is worthwhile noting that ConPath determines the relative orientations of all contigs using the orientations of the end reads.

Figure 2 shows the determination of the order and orientations of three contigs using two mate-pairs. In Figure 2(a), and reads determine the relative orientation of contigs and , and, in the same way, and reads determine the relative orientations of contigs and (see Figure 2(b)). The relative orientations of contigs and are determined by rotating the scaffold in Figure 2(b), as shown in Figure 2(c).

Figure 2: Determining the relative orientations of contigs using mate-pair information. (a): and reads determine the relative orientation of contigs and ; (b): and reads determine the relative orientations of contigs and ; and (c): the relative orientations of contigs , , and are determined by rotating the scaffold in Figure 2(b).

2.4. Estimation of the Gap Size between Contigs

Assuming all mate-pairs have a fixed size, the size of the gap between two adjacent contigs is determined by the sizes of the two contigs and the positions of the end reads of contigs.

Suppose that contig contains read and contig contains g read. Let Gap be the gap size between and . Let and be the start and end positions of read in , respectively, and let and be the start and end positions of read in , respectively. Considering all the possible directions of a mate-pair of two end reads, ConPath estimates the gap size as

b-U and g-U:Gap b-U and g-C: Gap b-C and g-U: Gap b-C and g-C: Gap

Figure 3 shows the procedure for estimating the gap size between contigs when and have and directions, respectively. The orientations of contigs and are set in the same direction. The length of part of the mate-pair library in contig and the length of part of the mate-pair library in contig are calculated. Finally, the gap size is calculated as

Figure 3: Estimation of the gap size between contigs when has direction and has direction . The gap size between and can be calculated as

2.5. Detection of Erroneous Contigs

One important feature of ConPath is the verification of a contig assembly by identifying erroneous contigs. We have defined 4 types of contig assembly errors to check the quality of a contig assembly.

Self-collision error
When the number of mate-pairs connecting two adjacent contigs is more than 2, and there is an inconsistency in determining the orientation of contigs with mate-pairs, the error is defined as a self-collision error, the most serious error type. If this error occurs, the contigs should be inspected manually one by one.

Mate-pair size error
When a mate-pair of an end read is contained in a contig, the real size of this mate-pair can be calculated. If the difference between the calculated and predefined sizes is larger than a threshold value, the error is defined as a mate-pair size error. This type of error is very critical to the contig assembly process.

Gap-size error
If the gap size between two contigs is a negative value, it indicates that the two contigs should be merged in the contig assembly process; this is defined as a gap size error.

Overlap error
After calculating the distances of all adjacent contigs, any two nonadjacent contigs can be overlapped due to the accumulation of errors in gap size estimations. This type of error is defined as an overlap error, which happens rarely and is not so critical.

Identifying error types is useful in verifying and correcting the final result of a contig assembly. If a contig has more than two types of errors, it is highly probable that a misassembled contig is present. ConPath assigns different colors to contigs by the number of error types, with nonerroneous contigs colored blue. When one contig has more than one error, ConPath assigns this contig a reddish color, with the intensity proportional to the number of error types. Therefore, we can check the quality of the final result of a contig assembly by simply inspecting the color information in the scaffold visualization window of ConPath.

2.6. Implementation

ConPath was implemented on a Windows XP system using Visual ++. It provides a user-friendly interface and shows visual and color-informative outputs, which can help analyze scaffolds both intuitively and informatively. ConPath provides dialogue windows for “mate-pair information”, “edge information”, “contig path”, and “invalid contigs" by automatically checking for the 4 types of errors. Scaffolds are displayed graphically in proportion to the real sizes of vertices and edges after aligning vertices and edges to avoid graphical collision, and the detailed information for each vertex and edge is shown on a pop-up window. ConPath can produce a large picture for all scaffolds by assembling separately printed module pictures. Figure 4 shows various viewers and dialogues of ConPath.

Figure 4: A set of snapshots of ConPath. ConPath provides a set of useful information, “mate-pair information”, “edge information”, “contig path”, and “invalid contigs” by checking for the 4 types of error.

3. Experiments and Discission

We tested ConPath using both artificial and real data. Artificial data were generated in two different versions: (randomly) and (uniformly). The version consisted of contigs of random sizes, whereas the version consisted of contigs of uniform size. In these artificial data experiments, ConPath showed very successful scaffold constructions using mate-pair information. From experiments with artificial data, ConPath made a reasonable scaffold construction in linear time.

ConPath worked very successfully and efficiently on real data sets, in sequencing the Mannheimia succiniciproducens and Vibro vulnificus genomes. ConPath verified the results of contig assembly by detecting misassembled contigs. Table 1 shows the mate-pair information in these real datasets. Four datasets were tested in sequencing the M. succiniciproducens genome, whereas one dataset was tested in sequencing the V. vulnificus genome, to verify the results of contig assembly. Table 2 shows these results. , , , and are the contig assembly results of the M. succiniciproducens genome and is the contig assembly result for the V. vulnificus genome. For the M. succiniciproducens genome, going from to increased the reliability of the contig assembly results.

Table 1: Mate-pair information in real test datasets. The proportion of mate-pair reads for V. vulnificus is about double that for M. succiniciproducens.

Table 2: Real test datasets. Four datasets for the M. succiniciproducens genome and one for the V. vulnificus genome were tested with ConPath. MP: mate-pair, MPIC: mate-pair in the same contigs.

We examined the edge number according to linking level (see Figure 5). ConPath was most successful at linking level 2 by minimizing the loss of edges.

Figure 5: Distribution of the number of edges according to linking level . ConPath constructed the best scaffolds at linking level 2 while minimizing edge loss.

Table 3 shows the detected errors in scaffold construction for the 5 datasets. Among the M. succiniciproducens datasets, had the most errors, whereas had no erroneous contigs. These results show that identifying the 4 types of errors for contigs is effective in verifying the result of contig assembly.

Table 3: Number of reported errors in scaffold construction for 5 dataset.

Figure 6 shows the constructed scaffolds at linking levels 2 and 3 for the dataset. Contig 93 is suspected of being erroneous because it has several erroneous contigs on both sides. ConPath showed that contig 93 was misassembled. The contig information dialogue box for contig 93 is shown in Figure 6(c).

Figure 6: An example of the detection of mis-assembled contigs. (a): Scaffolds for at linking level 2; (b): scaffolds for at linking level 3; (c): information on contig 93.

Table 4 shows a comparison of features of several scaffold analysis tools, including ConPath, Consed [9], Autofinish [10], and Bambus [12]. Compared with these other tools, ConPath has very good features for 5 criteria. Most importantly, ConPath helps users to intuitively verify the contig assembly by providing many visualization features and additional information to detect erroneous contigs.

Table 4: Comparison of ConPath with other scaffold tools.

4. Conclusion

A scaffold analyzer is a very important tool in genome sequencing, in that it can verify the results of contig assembly and to identify misassembled contigs. We have developed ConPath, a scaffold analyzer that exploits mate-pair information to construct scaffolds by ordering and orienting separate sequence contigs. ConPath provides various useful viewers and dialogue boxes for intuitive understanding. Using end read pairs of a fixed-sized mate-pair library, ConPath can determine the relative orientations of all contigs successfully, and estimate the gap size of each adjacent contig pair. We defined 4 types of errors to detect misassembly. ConPath was used successfully in sequencing several microbial genomes, including the M. succiniciproducens genome [13]. ConPath is, therefore, a useful scaffold analyzer to verify contig assembly by detecting erroneous contigs.

ConPath will doubtless improve as its algorithm becomes more correct and efficient, as well as through the development of additional features, such as primer design for the finishing step and a sequence read viewer.