 Research article
 Open Access
 Published:
Local alignment of generalized kbase encoded DNA sequence
BMC Bioinformatics volume 11, Article number: 347 (2010)
Abstract
Background
DNA sequence comparison is a wellstudied problem, in which two DNA sequences are compared using a weighted edit distance. Recent DNA sequencing technologies however observe an encoded form of the sequence, rather than each DNA base individually. The encoded DNA sequence may contain technical errors, and therefore encoded sequencing errors must be incorporated when comparing an encoded DNA sequence to a reference DNA sequence.
Results
Although twobase encoding is currently used in practice, many other encoding schemes are possible, whereby two ore more bases are encoded at a time. A generalized kbase encoding scheme is presented, whereby feasible higher order encodings are better able to differentiate errors in the encoded sequence from true DNA sequence variants. A generalized version of the previous twobase encoding DNA sequence comparison algorithm is used to compare a kbase encoded sequence to a DNA reference sequence. Finally, simulations are performed to evaluate the power, the false positive and false negative SNP discovery rates, and the performance time of kbase encoding compared to previous methods as well as to the standard DNA sequence comparison algorithm.
Conclusions
The novel generalized kbase encoding scheme and resulting local alignment algorithm permits the development of higher fidelity ligationbased next generation sequencing technology. This bioinformatic solution affords greater robustness to errors, as well as lower false SNP discovery rates, only at the cost of computational time.
Background
DNA sequence comparison is a well studied problem in biology and bioinformatics [1–4]. Recently, a new DNA sequencing technology (ABI SOLiD sequencing) has been developed which does not measure each base directly, but instead measures DNA bases in pairs in an encoded form [5–7]. This technology has the potential to have greater error tolerance by differentiating biological variants from sequencing errors. In this manner, Homer et al. (2009) previously developed an algorithm to compare an encoded DNA sequence to a target reference [8], with a similar method independently derived in Rumble et al. (2009) [9]. These two algorithms do not significantly differ and therefore the proposed algorithm is compared to the algorithm presented in Homer et al. (2009). This twobase encoding and resulting local DNA sequence comparison algorithm can be utilized with global search strategies [9–12] for wholegenome sequencing with nextgeneration sequencing technology [13].
The central advantage of the twobase encoding scheme is that the false discovery rate of a single nucleotide polymorphism (SNP) is reduced, since two specific adjacent errors are required to produce a SNP call. In fact, only onefourth of all adjacent errors would result in a false call. This significantly reduces the probability of falsely observing a SNP, with current machines exhibiting a color read error rate less than 5%. Nevertheless, the currently implemented twobase encoding is not the only possible encoding. Therefore a generalized kbase encoding scheme is presented, whereby k consecutive bases are simultaneously observed. The algorithm of Homer et al. (2009) is extended to solve the DNA sequence comparison problem of comparing a kbase encoded DNA sequence and a target reference DNA sequence. Intuitively, with greater k, the number of errors required to falsely discover a SNP also becomes greater, thus allowing machine errors to be accurately identified, and even corrected, while retaining sensitivity to detect real base changes. Simulations are performed to explore the improved power of higher order kbase encoding schemes, as well as the performance time when utilizing these encodings. These simulations explore the case for adapting kbase encoding schemes in ligationbased next generation sequencing technology.
Results and Discussion
Simulations were performed to explore the power and performance of kbase encoding as well as the kbase encoding local alignment algorithm (see Methods). Reads were simulated using sequences both with a uniform errorrate, as well as using sequences with an errorrate modeled after realworld data. In this discussion,"1base encoding" and "no encoding" are used interchangeably. The power to align a sequence with or without SNPs is defined as the fraction of reads where the read sequence is aligned to the reference with the same alignment score as if the sequence were aligned to the correct location (see Methods). The fraction of reads where the read is aligned to call a SNP, and where the original sequence had no SNP, defines the false positive SNP discovery rate. Similarly, the fraction of reads where the read is aligned not to call a SNP, and where the original sequence had a SNP, defines the false negative SNP discovery rate.
Plotted in Figure 1 is the power to align encoded sequences of varying length (25, 50, and 75 bp) with 02 SNPs or base substitutions given a fixed uniform error rate for k = 1...5 (see Methods). The power decreases similarly for all read lengths as the errorrate increases given a fixed number of SNPs. Furthermore, the power decreases substantially when the number of SNPs in the sequence is increased at a fixed errorrate and read length. The power of 1base encoding, observing each base directly, does not diminish as much as kbase encoded sequences (k > 1) when more SNPs are introduced. This is due to SNPs and observational errors being equivalent in the 1base encoding case. In these simulations, kbase encoding is more powerful when k > 2 for 0 SNPs, when k > 3 for 1 SNP, and when k > 4 for 2 SNPs and longer read lengths. It is important to note that in many cases when the alignment score between the best alignment and correct alignment differ, the decoded base sequences match. Therefore, the false positive and false negative SNP discovery rates are examined later.
To assess the power of kbase encoding utilizing realworld error rates, the color error (encoding error) rates were estimated from a previous run of an ABI SOLiD v2 sequencer (see Methods). The power of aligning such sequences was assessed in the presence of 02 SNPs for k = 1...5 (Table 1). For sequences with no SNPs, the power of kbase encoding increases as k increases. However in the context of an increasing number of SNPs and for k > 1 the power of kbase encoding is more ambiguous. For one SNP, k = 2 performs more poorly than no encoding, while k > 2 improves on the lower width encodings. For two SNPs, both k = 2 and k = 3 perform more poorly than no encoding, with only k > 3 having better power than no encoding. Thus, error correction is increased with greater k when no SNPs are present. However, if the goal is to find variants, a large enough k must chosen carefully to justify the penalty in performance.
The false positive SNP discovery rate is evaluated for 25, 50, and 75 basepair reads (Figure 2). With no encoding, SNPs and errors are not distinguishable, and therefore k = 1 is omitted from this discussion. As expected, the false positive SNP discovery rate decreases as k increases. Nevertheless, only above a five percent error rate does 2base encoding begin to find false SNPs, and at approximately ten percent error rate do all encodings considered begin to falsely discover SNPs. Assessed in Figure 3 is the false negative SNP discovery rate. Similarly to Figure 2, the false positive SNP discovery rate, the false negative SNP discovery rate decreases as k increases. For low error rates, both of the above metrics are either zero (for Figure 2) or are less than 20% (for Figure 3). Thus, the settings can be interpreted as being conservative, sacrificing power to find true SNPs for decreasing the false positive SNP discovery rate.
To illustrate the flexibility of kbase encoding to be tuned for specific scenarios, the power of 5 base encoding is examined when the score for a color substitution is varied, and for 25, 50, and 75 basepair reads with 0  2 SNPs. Intuitively, the various color substitution scores correspond to preferring a given number of color errors over a SNP and possibly fewer color errors (see Methods). As the color error score decreases, the encoding begins to prefer decoding with SNPs rather than with color errors. The various scoring schemes allow for a clear trade off in power to detect color errors over the power to detect SNPs (Figure 4). For example, the color substitution score of 25 allows the full correction of 50 bp reads with up to a 20% error rate. However, once a SNP is introduced it has zero power. Alternatively, the color substitution score of 200 finds SNPs in the low error case, but with higher error data the power to detect only the given SNP(s) is confounded as more SNPs are falsely detected. With zero SNPS, color scores of 25, 50, and 75 have almost perfect power.
The performance time of kbase encoding scales exponentially with increasing k (see Methods). This is confirmed by plotting the timing information from the previous uniform errorrate simulations (Table 2). Since the optimizations found in Homer et al. (2009) [10] are not pursued, increased variability was observed when the number of SNPs increased. However, an exponential increase in running time occurs as the encoding width (k) increases. This has practical implications, whereby producing empirical data that is kbase encoded is possible, but is computationally infeasible to decode as the number of shortreads is in the millions, if not billions for typical experiments.
Nevertheless, this exponential increase in running time could be significantly reduced at the cost of completeness by using methods initially adopted for protein similarity search and sequence comparison [14, 15]. A global search strategy is employed to put constraints on the possible search paths thereby significantly reducing the search space. Typically these constraints force the alignment path to proceed along diagonals in the dynamic programming matrix, reducing the dimensionality of the problem. These constraints are specifically used in the implementation of BFAST for both 1base and 2base encoded data [10]. The positions where the read and reference match during the BFAST's global search strategy are annotated and aggregated such that in the local alignment the read and reference must match (i.e. constraining the solution path to pass through a diagonal). In BFAST's case, the global search strategy searches over an 2base encoded reference for k = 2 constraining the color transitions. The constraints employed by BFAST could be applied for k greater than two, yielding significant but unknown speed improvement.
Conclusions
The generalized kbase encoding scheme and resulting local alignment algorithm presented here have the ability to more powerfully differentiate between encoded sequencing errors and true DNA variants. These schemes can be used in practice to tolerate high error rates in the raw data. Alternatively, the perbase accuracy of sequencing can be improved. The goal in most sequencing projects is to sensitively and specifically detect variants. The technology and encoding scheme must not only have sufficient power to detect variants but to also not overwhelm the true variants with false variants. The results demonstrate that higher encoding schemes not only improve the power of detecting variants, but also significantly reduce the false positive SNP discovery rate. Having multiple observations of a specific variant or genomic coordinate (higher coverage) is in most cases able to overcome sequencing error. Nevertheless, these encoding schemes could allow low coverage data to accurately detect variants. Furthermore, for cancer specific studies, where the sample may be a heterogeneous population of cells, these encodings could reduce the minimum detection level (allele frequency) in the cancer cell population as the fewer observations can be more confidently trusted.
Currently a twobase encoding system is used by ABI SOLiD sequencing technology. Some other nextgeneration sequencing technologies could also adopt an encoding system to improve their performance and accuracy. Furthermore, algorithms that perform multiple sequence alignment or local reassembly could also utilize the power of the encoding scheme presented here. It is interesting to note that error correction utilizing encoded DNA sequence could be performed if single bases or sets of bases were observed more than once. Utilizing various encoding schemes, this error correction would necessarily not rely on a target DNA reference comparison, thereby eliminating the expensive exponential increase in time for higher order encodings (larger k). Future investigation of such prealignment error correction schemes and algorithms is intended.
Methods
Generalized kbase encoding
Given an kbase encoded DNA sequence c = c_{1}, ..., c_{ n }, it is the goal of the proposed algorithm to minimize the edit distance between c and some regular DNA sequence y = y_{1}, ..., y_{ m }given a set of valid edit operators Σ. The DNA alphabet is assumed to be Λ = {A, C, G, T} and the valid edit operators include a color substitution, base substitution, a deletion, and an insertion. Similar to Homer et al. (2009), the color substitution operator is required to be applied before applying the insertion, deletion, or base substitution operators. Each operator is assigned a score, with ∏(C_{1}, C_{2}) and Δ(B_{1}, B_{2}) corresponding to the color substitution scoring and base substitution scoring functions respectively. To model insertions and deletions, affine gap penalties are used whereby a score of ρ is applied for the first insertion (or deletion), with ϵ applied for any consecutive insertion (or deletion) that extends the insertion (or deletion). It is assumed that for any bases B_{1} ≠ B_{2} and for any colors C_{1} ≠ C_{2} that 0 ≤ Δ(B_{1}, B_{1}), Δ(B_{1}, B_{2}) < 0, 0 ≤ ∏(C_{1}, C_{1}), ∏(C_{1}, C_{2}) < 0, ρ < ϵ < 0. These scoring assumptions penalize edits that change the encoded or decoded DNA sequence relative to the reference, thereby ensuring their similarity.
To illustrate the encoding and decoding method used by this technology, let x = x_{1}, ..., x_{ n }be a DNA sequence. To encode a DNA sequence, the function Φ^{k}(B_{1}, ..., B_{ k }) is defined to return a color C_{ k }using the bases B_{1}, ..., B_{ k }, where B_{i1}occurs before B_{ i }in the sequence. For example, to encode the DNA sequence x = x_{1}, ..., x_{ n }, first a known start adaptor p = p_{1}, ..., p_{k1}∈ Λ^{k1}is assumed. Next, the function Φ^{k}is iteratively applied to the concatenation of p and x. c In this case, c_{1} = Φ^{k}(p_{1}, ..., p_{k1}, x_{1}), c_{2} = Φ^{k}(p_{2}, ..., p_{k1}, x_{1}, x_{2}), ..., c_{ n }= Φ^{k}(x_{nk+1}, x_{ n }). The adaptor sequence p is known in practice and is used in the physical chemistry of the sequencer (for k = 2), not the DNA sequence in question [5–7].
The encoding function Φ^{k}(B_{1}, ..., B_{ k }) transforms each base B_{ i }into an integer representation (i.e. A = 0, C = 1, G = 2, T = 3), sums the integer values, and returns the result modulo four. Let δ return the integer representation of a base as described above, then Φ^{k}(B_{1}, ..., B_{ k }) = mod Λ. Modulo four is chosen since four colors are used in current technologies. The properties of the modulofourspecific encoding are discussed after how a kbase encoded sequence is decoded.
To decode an encoded sequence, the function Γ^{k}(B_{1}, ..., B_{k1}, C) is defined to return the decoded base B_{ k }using the encoded color C and the previous bases B_{1}, ...B_{k1}. To compute Γ^{k}(B_{1}, ..., B_{k1}, C), B must be solved for in the equation mod Λ, which is easily solved. For example, to decode the encoded sequence c = c_{1}, ..., c_{ n }with a known start adaptor p = p_{1}, ...p_{k1}∈ Λ^{k1}, Γ^{k}is iteratively used. The decoded sequence will be x_{1} = Γ^{k}(p_{1}, ..., p_{k1}, c_{1}), x_{2} = Γ^{k}(p_{1}, ..., p_{k2}, x_{1}, c_{2}), ..., x_{ n }= Γ^{k}(x_{nk+1}, x_{n1}, c_{ n }). Without the start adaptor p, there would be Λ^{k1}possible decodings of the encoded sequence.
This encoding function has two useful properties. First, if one base in x is changed to obtain a new DNA sequence x', then the new k colors that encode the changed base in x' will differ from the corresponding k colors in x. For example, if k = 5 then changing one base in x to obtain new DNA sequence x' will cause there to be 5 color differences between the encoded version of x and the encoded version of x'. A second useful property is if one color in the encoded version of x is changed to obtain a new encoded version, say c', then every base in the decoded version of c' that occurs after the changed color will be different from the corresponding bases in x. The first property defines the signature of base substitutions in the encoded sequence, which becomes pronounced as k increases. The second property tells us that an encoding error will modify all bases after the encoding error. Intuitively, one can simplify by observing that for any base substitution there exists a set of k consecutive errors that can be applied to achieve the same base substitution, and therefore if variants are being searched for (base substitutions in particular) then constraints should be placed on the scoring system to prefer calling base substitutions rather than color substitutions when comparing to a reference. An example of such a constraint is given that removes the above ambiguity. Suppose there exists a subsequence of the encoded read , such that they all encode a base . Next, consider the reference base B_{ i }≠ and the k "colors" that encode B_{ i }: C_{ i }, ..., C_{i+k1}. Let the i th DNA base be such that encode B_{ i }. The following constraint is made to prefer a base change and k color matches to a base match and k consecutive color mismatches:
In this case, it is assumed that C_{ j }≠ and B_{ i }≠ (∀i ≤ j ≤ i + k  1). Numerous other constraints based on realworld requirements are possible but not explored here.
The Algorithm
Suppose that a color sequence c = c_{1}, ..., c_{ n }with a known adaptor p ∈ Λ^{k1}is to be aligned to a reference sequence y = y_{1}, ..., y_{ m }. To search over all possible base substitution, base insertion, base deletions, and color substitutions, define a recursive formula that is the repeated calculation in the dynamic programming algorithm.
Intuitively, Equation 2 is filling in an n by m matrix, with each cell in the matrix containing 3 × Λ^{k1}subcells. It is interesting to observe for k > 2 when computing and that 2 considers only previous subcells that are consistent with the current subcell. In other words, the first k  2 bases of σ (the current subcell) must correspond to the last k  2 bases of ϕ (the previous subcell). In this formula, the h subcells represent bases present in y but not in x, while v subcells represent bases present in x but not in y. The s subcells represents a base x_{ i }aligning to a base y_{ i }in the reference sequence y.
An alignment that begins or ends with a deletion is ignored, since a sequence must span the breakpoint for the deletion to be observable (with respect to x). This is a valid assumption when x is the observed sequence, and y is a fixed reference. An insertion followed by a deletion (or vice versa) is ignored since this is rare for short DNA sequences, although to consider such an event would require minimal changes to the above formula.
If the color match scores are the same (∀i ≠ j, ∏(c_{ i }, c_{ i }) = ∏(c_{ j }, c_{ j })) and all color mismatch scores are the same (∀i ≠ j, k, ≠ l, ∏(c_{ i }, c_{ j }) = ∏(c_{ k }, c_{ l }), then Equation 2 can be simplified. The recursive rule for the term becomes:
This modification forces any color substitution to be at the beginning or end of any inserted bases in x and can reduce the complexity of the algorithm dramatically. The intuition behind this simplification stems from not having any reference bases to which to compare the inserted bases. This forces the maximum score path through the insertion to have no color errors
Various initializations are possible, and the alignment of the entire encoded DNA sequence x to some subsequence of y is presented here. Therefore, the initialization becomes for , if σ = Γ^{k}(p, c_{ i }) and otherwise, and for if σ = Γ^{k}(ϕ, c_{ i }), so that the local alignment spans the entire encoded sequence and insertions are allowed at the beginning of any alignment. Notice that if there were any color errors within the beginning an insertion, they are aligned such that they occur at the end of the insertion. for j ≥ 0 is initialized so that the alignment does not begin with a deletion. The remaining initializations are: for j ≥ 0 σ = p and σ ∈ Λ^{k1}, and if σ = p, otherwise, for j ≥ 0 and σ ∈ Λ^{k1}. These initializations enforce that the starting adaptor is p. To find the optimal local alignment, the cells and are searched over for a cell with maximum score, again ignoring the case where the alignment ends with a deletion. Backtracking is used to recover maximum scoring alignment.
This algorithm is in fact finding the shortest path on a graph with the nodes defined by the subcells of the matrix, and the edges weighted and defined by the recursive rules. To analyze the time complexity, it is observed that given the kbase encoding scheme for each subcell of type h, v, and s there are Λ^{k1}subcells. For each subcell it is required to calculate the maximum over two values. For each and subcell it is required to calculate the maximum over 3 × Λ values. Therefore for each cell, various maximum must be computed over 2 × Λ^{k1}+ Λ^{k1}× 3 × Λ + Λ^{k1}× 3 × Λ = 2 × Λ^{k1}(1 + 3 × Λ). In practice, Λ = 4 and therefore various maxima must be computed over 26 × 4^{k1}values. From this analysis, it is clear that the running time of this algorithm is O(nmΛ^{k}), which unfortunately scales exponentially with respect to the length of the encoding k.
Simulations
Simulations were performed to assess the power and performance of kbase encoding, for k = 1...5. Sets of 10, 000 test sequences were randomly sampled from E. Coli (DH10B, NC_010473, CP000948). All sequences in a given set had a fixed read length (25, 50, 75), a fixed errorrate (0, 0.01, ..., 0.2), and a fixed number of SNPs (0, 1, 2). For the case of the 1base encoding (k = 1), the standard SmithWaterman algorithm was used, where errors were modeled as base changes, requiring that SNPs and base errors do not cooccur. For the case of kbase encoding with k > 1 errors were modeled as color substitutions (encoding errors). Similar to Homer et al. (2009), an alignment is defined to be accurate or correct if the returned alignment has the same score (or likelihood when the scores represent log likelihoods) as the true alignment, which is known by the nature of these simulations.
To allow for insertions and deletions, the original sequence is used (before applying errors and variants) with an additional 10 bp before and after as the reference or target DNA sequence. In accordance with Equation 1, ϵ =  50, ρ =  175, ∏(C_{1}, C_{2}) = 125 (C_{1} ≠ C_{2}), ∏(C_{1}, C_{1}) = 0, Δ(B_{1}, B_{2}) =  150(B_{1} ≠ B_{2}) and Δ(B_{1}, B_{1}) = 50. Due to these initializations, the optimization in Equation 3 is able to be performed. To model realworld errorrates, the simulated errorrates are learned from a run of an ABI SOLiD sequencer (50 base pairs), utilizing the aligned reads to calculate the 2base encoding error, which is inherently dependent on the decoding algorithm used Homer et al. (2009). The errorrate was not uniform by sequencing position, therefore producing a colorerrorrate for each position in the 50 color sequence reads. The observed error rate for each sequencing position was: 0.014, 0.005, 0.006, 0.007, 0.006, 0.006, 0.006, 0.008, 0.008, 0.008, 0.007, 0.006, 0.009, 0.009, 0.009, 0.009, 0.008, 0.015, 0.015, 0.012, 0.012, 0.011, 0.021, 0.021, 0.018, 0.019, 0.014, 0.037, 0.033, 0.031, 0.029, 0.022, 0.055, 0.052, 0.051, 0.043, 0.036, 0.087, 0.084, 0.076, 0.071, 0.060, 0.125, 0.118, 0.118, 0.108, 0.092, 0.179, 0.175, 0.184. To evaluate various scoring schemes of 5base encodings, simulations of only 1, 000 test sequences were used due to running time limitations. For these evaluations a dual quadcore Intel Xeon E5420 machine at 2.5 GHz, with 32 GB of RAM and 2TB of RAID 0 disk space, was used, although the actual hardware requirements of the algorithm itself beyond CPU power are negligible relative to any modern computer.
Scoring constraints 5base encoding
Various scoring schemes were evaluated for 5base encoding. For notational convenience, for all colors C_{1} ≠ C_{2} and bases B_{1} ≠ B_{2} let CE = ∏(C_{1}, C_{2}), CM = ∏(C_{1}, C_{1}), BE = Δ(B_{1}, B_{2}), and BM = Δ(B_{1}, B_{1}). Consider the scoring scenarios that satisfy one of the following constraints:

1.
5CE + BM > 5CM + BE (25)

2.
5CE + BM < 5CM + BE and 4CE + CM + BM > CE + 4CM + BE (50)

3.
4CE + CM + BM < CE + 4CM + BE and 3CE+2CM+BM > 2CE+3CM+BE (75 and 150)

4.
3CE +2CM +BM < 2CE +3CM +BE and 2CE+3CM+BM > 3CE+2CM+BE (200)

5.
2CE +3CM +BM < 3CE +2CM +BE and 1CE + 4CM + BM > 4CE + CM + BE

6.
CE + 4CM + BM < 4CE + CM + BE
Intuitively, these scenarios try to decide if a given set of color errors should be preferred if they can be explained by a SNP and possibly other color errors. For example, the first scenario always prefers calling color errors over anything that can be explained by a SNP. The second scenario will prefer to explain the encoding with a SNP if it results in no color errors, but does not prefer to explain the encoding with a SNP if it is accompanied by any color errors. In the extreme, the last scenario would prefer to explain all color errors as a combination of a SNP and possibly color errors.
Nevertheless, given the assumptions that BM ≥ 0, CM ≥ 0, BE < 0, and CE < 0, it is observed that scenarios 5  6 are not possible. For example, constraint is equivalent to the following constraint: CM + BM < CE + BE and 4CE + BM > 3CE + BE. Intuitively, our assumptions prefer to penalize color errors and nucleotide variants, rewarding color matches and nucleotide matches, thereby making the leftside of the above constraint without a solution. If scenarios 56 are truly desired, instead of considering color errors, color matches, base errors, and base matches separately, a joint function could be considered conditional on the various combinations of events over the k base and color window. This would also allow for the incorporation of any specific experimental bias for the combinations of events, or even specific bases.
In the above constraints color error scores are given that satisfy the constraints given the previously defined base match, base substitution, and color match scores. The score 150 is also included, which was previously used, to illustrate that there is flexibility even within these constraints to tune the scoring scheme.
References
 1.
Hamming R: Error Detecting and Error Correcting Codes. Bell System Technical Journal 1950, 26(2):147–160.
 2.
Gotoh O: An improved algorithm for matching biological sequences. J Mol Biol 1982, 162: 705–708. 10.1016/00222836(82)903989
 3.
Needleman S, Wunsch C: A general method applicable to the search for similarities in the amino acid sequence of two proteins. J Mol Biol 1970, 48: 443–453. 10.1016/00222836(70)900574
 4.
Smith T, Waterman M: Identification of common molecular subsequences. J Mol Biol 1981, 147: 195–197. 10.1016/00222836(81)900875
 5.
ABI: Principles of DiBase Sequencing and the Advantages of Color Space Analysis in the SOLiD System. In Tech. Rep. 139AP10–01. Applied Biosystems Incorporated; 2008.
 6.
ABI: A Theoretical Understanding of 2 Base Color Codes and Its Application to Annotation, Error Detection, and Error Correction. In Tech. Rep. 139WP01–01. Applied Biosystems Incorporated; 2008.
 7.
Smith D, Quinlan A, Peckham H, Makowsky K, Tao W, Woolf B, Shen L, Donahue W, Tusneem N, Stromberg M, Stewart D, Zhang L, Ranade S, Warner J, Lee C, Coleman B, Zhang Z, McLaughlin S, Malek J, Sorenson J, Blanchard A, Chapman J, Hillman D, Chen F, Rokhsar D, McKernan K, Jeffries T, Marth G, Richardson P: Rapid wholegenome mutational profiling using nextgeneration sequencing technologies. Genome Res 2008, 18: 1638–1642. 10.1101/gr.077776.108
 8.
Homer N, Merriman B, Nelson S: Local alignment of twobase encoded DNA sequence. BMC Bioinformatics 2009, 10: 175. 10.1186/1471210510175
 9.
Rumble SM, Lacroute P, Dalca AV, Fiume M, Sidow A, Brudno M: SHRiMP: accurate mapping of short colorspace reads. PLoS Comput Biol 2009, 5: e1000386. 10.1371/journal.pcbi.1000386
 10.
Homer N, Merriman B, Nelson SF: BFAST: an alignment tool for large scale genome resequencing. PLoS ONE 2009, 4: e7767. 10.1371/journal.pone.0007767
 11.
Kent W: BLATthe BLASTlike alignment tool. Genome Res 2002, 12: 656–664.
 12.
Li H, Ruan J, Durbin R: Mapping short DNA sequencing reads and calling variants using mapping quality scores. Genome Res 2008, 18: 1851–1858. 10.1101/gr.078212.108
 13.
Clark MJ, Homer N, O'Connor BD, Chen Z, Eskin A, Lee H, Merriman B, Nelson SF: U87MG decoded: the genomic sequence of a cytogenetically aberrant human cancer cell line. PLoS Genet 2010, 6: e1000832. 10.1371/journal.pgen.1000832
 14.
Lipman DJ, Pearson WR: Rapid and sensitive protein similarity searches. Science 1985, 227: 1435–1441. 10.1126/science.2983426
 15.
Pearson WR: Rapid and sensitive sequence comparison with FASTP and FASTA. Meth Enzymol 1990, 183: 63–98. full_text
Acknowledgements
This research was partially supported by University of California Systemwide Biotechnology Research and Education Program GREAT Training Grant 200710 (to NH), the NIH Neuroscience Microarray Consortium (U24NS052108), and a grant from the NIMH (R01 MH071852). We would also like to thank members of the Nelson Lab: Zugen Chen, Hane Lee, Bret Harry, Jordan Mendler, Brian O'Connor for input and computational infrastructure support. Finally, we would like to thank the anonymous reviewers for their very insightful and helpful comments and suggestions.
Author information
Additional information
Authors' contributions
BM and NH conceived of kbase encoding. NH conceived of and implemented the algorithm, and performed the analyses. BM and SFN advised on the development and analysis of the method, and producing the manuscript. All authors read and approved the final manuscript.
Authors’ original submitted files for images
Below are the links to the authors’ original submitted files for images.
Rights and permissions
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
About this article
Received
Accepted
Published
DOI
Keywords
 Single Nucleotide Polymorphism
 Encode Scheme
 Base Substitution
 Color Error
 Percent Error Rate