Home | About | Journals | Submit | Contact Us | Français |

**|**BMC Res Notes**|**v.5; 2012**|**PMC3483159

Formats

Article sections

- Abstract
- Background
- Findings
- Conclusion
- Availability and requirements
- Competing interests
- Authors’ contributions
- References

Authors

Related links

BMC Res Notes. 2012; 5: 286.

Published online 2012 June 12. doi: 10.1186/1756-0500-5-286

PMCID: PMC3483159

Yonil Park: vog.hin.mln.ibcn@krap; Sergey Sheetlin: vog.hin.mln.ibcn@nilteehs; Ning Ma: vog.hin.mln.ibcn@gninam; Thomas L Madden: vog.hin.mln.ibcn@neddam; John L Spouge: vog.hin.mln.ibcn@eguops

Received 2012 March 30; Accepted 2012 May 16.

Copyright ©2012 Park et al.; licensee 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.

This article has been cited by other articles in PMC.

Local alignment programs often calculate the probability that a match occurred by chance. The calculation of this probability may require a “finite-size” correction to the lengths of the sequences, as an alignment that starts near the end of either sequence may run out of sequence before achieving a significant score.

We present an improved finite-size correction that considers the distribution of sequence lengths rather than simply the corresponding means. This approach improves sensitivity and avoids substituting an *ad hoc* length for short sequences that can underestimate the significance of a match. We use a test set derived from ASTRAL to show improved ROC scores, especially for shorter sequences.

The new finite-size correction improves the calculation of probabilities for a local alignment. It is now used in the BLAST+ package and at the NCBI BLAST web site ( http://blast.ncbi.nlm.nih.gov).

Local alignments are an essential tool for biologists and often provide the first information about the function of an unknown nucleotide or protein sequence. An important question concerns the relationship of the score of a local alignment with the probability that the alignment occurred by chance.
[Karlin and Altschul 1] developed an asymptotic theory for local alignments, assuming that no gaps are permitted. For two random sequences **I** and **J** of lengths *m* and *n*, respectively, the resulting distribution of the optimal alignment score
$\widehat{M}$ approximates a Gumbel distribution
[2]

$$\mathbb{P}\left\{\widehat{M}>y\right\}\approx 1-exp\left(-kmn{e}^{-\lambda y}\right)\text{.}$$

(1)

The two statistical parameters in Equation (1) are *λ*, the scale parameter, and *k*, the pre-factor.

Several authors
[3-12] extended this framework to local alignments with gaps and showed that the Gumbel distribution from Equation (1) is still valid, though different values for *λ* and *k* are required.
[Altschul 13] discussed the need for a “finite-size correction” to the lengths *m* and *n* to improve the accuracy of Equation (1). The resulting statistics are an integral part of the Basic Local Alignment Search Tool (BLAST)
[14].

The following presentation emphasizes intuition over mathematical formality, to explain how the finite-size correction can account for the finite sequence lengths *m* and *n* to improve the accuracy of Equation (1). Let us begin with an optimal local alignment, which starts from score 0 and requires a non-zero sequence length within both **I** and **J**, before it achieves score *y*. Let *L*_{I} (*y*) (*L*_{J} (*y*)) be the required random lengths within both **I** (**J**), and let
${l}_{I}\left(y\right)=\mathbb{E}\left\{{L}_{I}\left(y\right)\right\}$ (
${l}_{J}\left(y\right)=\mathbb{E}\left\{{L}_{J}\left(y\right)\right\}$) be the corresponding means. The main idea is that the optimal local alignment cannot start anywhere along the full length *m* (*n*) of sequence **I** (**J**), because there might be insufficient sequence to permit it to achieve the score *y* (Figure
(Figure1).1). The finite-size correction described in
[13] and used in BLAST therefore replaced the area *mn* of the alignment matrix for Equation (1) by

$$\left[m-{l}_{I}\left(y\right)\right]\left[n-{l}_{J}\left(y\right)\right]\text{.}$$

(2)

Equation (2) approximates the area within the alignment matrix where the optimal local alignment can start and on average still have enough space to exceed the score *y*. If *m*<*l*_{I} (*y*) or *n*<*l*_{J} (*y*), however, the resulting value in Equation (2) might become negative. The BLAST code for the old finite-size correction therefore set the corrected sequence length to an *ad hoc* value (typically 1). For very short query or database sequences, the *ad hoc* correction could underestimate the significance of an alignment by many orders of magnitude.

The purpose of this note is to present a new finite-size correction formula for the BLAST statistics. It avoids the *ad hoc* correction and improves on them by considering the (approximately normal) distributions of the random lengths *L*_{I} (*y*) and *L*_{J} (*y*) explicitly, and not just the corresponding means *l*_{I} (*y*) and *l*_{J} (*y*). We demonstrate below that the new finite-size correction is better than the older one, both in theory and in practice. All BLAST+ protein-protein applications (i.e., BLASTP, BLASTX) use the new finite-size correction by default, starting with version 2.2.26.

As with the old finite-size correction, the expectation ${l}_{I}\left(y\right)=\mathbb{E}{L}_{I}\left(y\right)$ is approximated linearly:

$${l}_{I}\left(y\right)={a}_{I}y+{b}_{I}\text{.}$$

(3)

Most practical scoring systems are symmetric, with *s*(*A*, *B*)=*s*(*B*, *A*) for any two letters *A* and *B*, and for a symmetric scoring matrix and symmetric sequence compositions, expectations corresponding to **I** and **J** are the same, e.g., *l*_{I} (*y*)=*l*_{J} (*y*)=*l*(*y*). For asymmetric scoring systems or asymmetric sequence compositions, however, the variates *L*_{I} (*y*) and *L*_{J} (*y*) can have different distributions, so the following retains the subscripts **I** and **J**.

The new finite-size correction replaces *mn* in Equation (1) by

$$\mathbb{E}\left\{{\left[m-{L}_{I}\left(y\right)\right]}^{+}{\left[n-{L}_{J}\left(y\right)\right]}^{+}\right\}\text{,}$$

(4)

where x^{+}=max{*x*,0}. Rather than taking the expectation of*L*_{I} (*y*) and *L*_{J} (*y*) as in Equation (2), Equation (4) is the expected area within the alignment rectangle where an optimal local alignment can start and have enough random sequence length to reach the score *y* (Figure
(Figure11).

The practical computation of Equation (4) approximates the distribution of (*L*_{I}(*y*),*L*_{J} (*y*)) with a bivariate normal distribution, with means
${l}_{I}\left(y\right)=\mathbb{E}{L}_{I}\left(y\right)$and
${l}_{J}\left(y\right)=\mathbb{E}{L}_{J}\left(y\right)$,variances var *L*_{I} (*y*)=*v*_{I} (*y*) and var *L*_{J} (*y*)=*v*_{J} (*y*), and covariance cov (*L*_{I}(*y*), *L*_{J} (*y*))=*c*(*y*), all assumed to be linear in the score *y*, i.e.,

$$\begin{array}{c}{l}_{I}\left(y\right)={a}_{I}y+{b}_{I},{l}_{J}\left(y\right)={a}_{J}y+{b}_{J}\text{,}\phantom{\rule{0.25em}{0ex}}\\ \phantom{\rule{0.25em}{0ex}}{v}_{I}\left(y\right)={\alpha}_{I}y+{\beta}_{I},{v}_{J}\left(y\right)={\alpha}_{J}y+{\beta}_{J},\\ \phantom{\rule{0.25em}{0ex}}c\left(y\right)=\sigma y+\tau \text{.}\end{array}$$

(5)

The estimation of the parameters *a*_{I}*a*_{J}*α*_{I}*α*_{J} and *σ* has mathematical depth and involves many unproved speculations, but involves a heuristic modeling of a random sequence alignment with Markov additive processes
[15], ultimately with use of the renewal-reward theorem. The Appendix presents formulas for computing *a*_{I}*a*_{J}*α*_{I}*α*_{J} and *σ*.

BLAST *p*-values are relatively insensitive to the values of the intercepts *b*_{I}*b*_{J}*β*_{I}*β*_{J}, and *τ*, so the practical computation approximates them, as follows. Let *a*_{u} (*α*_{u}) be the value of *a*_{I} (*α*_{I}) for ungapped alignment. The mathematical theories for random walks and for renewals yield analytic formulas for *a*_{u} and *α*_{u}[16]. For an ungapped optimal alignment, the alignment length required to exceed the score *y* is the same within the sequences **I** and **J**, because it lacks gaps. Thus, *a*_{u} and *α*_{u} do not depend on the sequence (**I** or **J**) under consideration, so they contain no subscripts **I** or **J**. In a gapped alignment, let a gap of length 1 incurs a penalty *G*. The following uncontrolled approximations hold
[17]:

$$\begin{array}{c}{b}_{I}=2G\left({a}_{u}-{a}_{I}\right),{b}_{J}=2G\left({a}_{u}-{a}_{J}\right)\\ {\beta}_{I}=2G\left({\alpha}_{u}-{\alpha}_{I}\right),{\beta}_{J}=2G\left({\alpha}_{u}-{\alpha}_{J}\right)\\ \tau =2G\left({\alpha}_{u}-\sigma \right)\text{.}\end{array}$$

(6)

Under the normal approximation, routine computation shows that Equation (4) is approximately

$$\begin{array}{l}\mathbb{E}\left({\left[m-{L}_{I}\left(y\right)\right]}^{+}{\left[n-{L}_{J}\left(y\right)\right]}^{+}\right)\approx \hfill \\ \left[\left(m-{l}_{I}\left(y\right)\right)\mathbb{P}\left(X\le {m}_{\Phi}\right)-\sqrt{{v}_{I}\left(y\right)}\mathbb{E}\left(X;X\le {m}_{\Phi}\right)\right]\times \hfill \\ \left[\left(n-{l}_{J}\left(y\right)\right)\mathbb{P}\left(X\le {n}_{\Phi}\right)-\sqrt{{v}_{J}\left(y\right)}\mathbb{E}\left(X;X\le {n}_{\Phi}\right)\right]+\hfill \\ c\left(y\right)\mathbb{P}\left(X\le {m}_{\Phi}\right)\mathbb{P}\left(X\le {n}_{\Phi}\right)\hfill \end{array}\text{,}$$

(7)

where
${m}_{\Phi}:=\left[m-{l}_{I}\left(y\right)\right]/\sqrt{{v}_{I}\left(y\right)}$,
${n}_{\Phi}:=\left(n-{l}_{J}\left(y\right)\right)/\sqrt{{v}_{J}\left(y\right)}$, and *X* is a standard normal variate. The final product
$\mathbb{P}\left(X\le {m}_{\Phi}\right)\mathbb{P}\left(X\le {n}_{\Phi}\right)$ is an uncontrolled independence approximation for the bivariate normal distribution.

We compared *p*-values for the new finite-size correction with those for the old finite-size correction using the BLOSUM62 scoring matrix and affine gap penalty 11+*g*. Hartmann used a rare-event simulation method to compute the local alignment score distribution for ranges that included small *p*-values like *p*=10^{−50}[18], thereby producing a theoretical standard for small *p*-values.

Figure
Figure22 plots relative errors in logarithmic scale against true *p*-values for equal sequence lengths *m*=*n*=40, 100, 200, and 400. Using Hartmann’s theoretical standard, the new finite-size correction outperforms as the *p*-value decreases, sometimes by orders of magnitude.

We evaluated the performance of the new finite-size correction using the ASTRAL SCOP 40 subset [19] of release 1.75 of the Structural Classification of Proteins (SCOP) [20] database. We sorted the SCOP domains by lexicographic order and used the even numbered sequences as our query set, but removed any query that was the sole member of the superfamily in ASTRAL 40. For a given query sequence, we considered any database sequence belonging to the same SCOP superfamily as a true positive, and any database sequence belonging to a different SCOP fold as a false positive. Following [21], in the retrieval list for each query, we censored all sequences belonging to the same fold but different superfamily, so those sequences contributed neither true or false positives to the retrieval.

We report the performance in terms of the Receiver Operator Characteristics (ROC). Specifically, we report the ROC_{n} score, which is obtained by pooling the results of all queries, ordering them by expect value, but only keeping results up the *n*-th false positive
[21]. The expect value for the database search was obtained from the pairwise *p*-values using a length-proportional correction that takes the ratio of the database length to the target sequence length into account
[13].

As discussed above, the new finite-size correction should show the greatest improvement for short sequences. Therefore, we also produced ROC_{n} scores for different subsets of the SCOP database. One database subset has sequences shorter than the 25th percentile length (95 residues), and another has sequences shorter than the 50th percentile length (137 residues).

Table
Table11 presents ROC_{n} scores for the full database as well as the two subsets described above. These scores have an average of one false positive per query (4852), a threshold found useful in other studies (Altschul SF, private communication). The ROC-4852 scores for the full database demonstrate a small improvement of the new finite-size correction over the older one. The subsets show a more impressive improvement. For the 50^{th} percentile subset, the ROC-4852 score improves by 9%. For the 25^{th} percentile subset, the ROC-4852 score shows a 13% improvement. In the 25^{th} percentile subset, the new finite-size correction produces roughly 12% more true positives overall at 4852 false positives than the old finite-size correction (Figure
(Figure3).3). These results confirm our expectation that the new finite-size correction will display greatest improvement in retrieval for short sequences.

Retrieval accuracy for different subsets of SCOP database with the new and old finite-size correction

To assess the significance of this improvement on BLAST searches, one may look to the length distribution of sequences in a heavily used protein BLAST database. The non-redundant (“nr”) database is the default protein database at the NCBI BLAST web site. Of the sequences in the nr database, 11% are 95 residues or shorter; and 21%, 137 residues or shorter. The new finite-size correction improves the retrieval accuracy for a noticeable fraction of the proteins in the nr database.

We have described a new finite-size correction. The new correction has a more rigorous derivation than the current finite-size correction and avoids the use of an *ad hoc* value for short sequences. We have tested the retrieval accuracy of the new finite-size correction on the gold standard SCOP set, and have shown that the improvement is most important for short sequences. This correction has been made part of the BLAST+ protein-protein applications (e.g., BLASTP, BLASTX) as well as at the NCBI BLAST web site. In the future, we plan to implement this correction for nucleotide-nucleotide comparisons.

Project Name: BLAST Statistical Parameters

Project home page: http://www.ncbi.nlm.nih.gov/CBBresearch/Spouge/html_ncbi/html/blast/

Operating systems: Windows, MacOSX, LINUX, UNIX

Programming language: C++

License: Public Domain (see http://www.ncbi.nlm.nih.gov/books/NBK22952/)

Any restrictions to use by non-academics: None

The authors declare that they have no competing interests.

YP, TM and JS drafted the manuscript. YP designed the *p*-value evaluation method. SS implemented the new finite-size correction. NM integrated the correction into the BLAST+ code, ran tests, and calculated the ROC scores. JS devised the new finite-size correction. YP and SS are equal contribution first authors for this article. TLM and JLS are equal contribution last authors for this article. All authors read and approved the final manuscript.

We thank Greg Boratyn for help in running the accuracy evaluations with the SCOP set. This research was supported by the intramural research program of the NIH, National Library of Medicine.

Let
${\mathbb{Z}}_{+}=\{0,1,\dots \}$. Consider two semi-infinite random sequences, yielding alignment scores *S*_{i,j} at each vertex
$\left(i,j\right)\in {\mathbb{Z}}_{+}^{2}$ within their alignment graph. Define the edge maximum score
${E}_{n}=max\left\{{max}_{0\le i\le n}{S}_{i,n},{max}_{0\le j\le n}{S}_{n,j}\right\}$. Let
${\kappa}_{0}={E}_{{\kappa}_{0}}=0$ and
${\kappa}_{i}=inf\left\{n:n>{\kappa}_{i-1},{E}_{n}>{E}_{{\kappa}_{i-1}}\right\}$ for *i*≥1. We call *κ*_{i} the *i*^{th} SALE (strict ascending ladder epoch) and
${E}_{{\kappa}_{i}}$ the *i*^{th} SALE score. Let
$\Delta {E}_{i}^{}:={E}_{{\kappa}_{i}}^{}-{E}_{{\kappa}_{i-1}}^{}$, the increment between the (*i*−1)^{th} and *i*^{th} SALE scores.

Let
${L}_{I}\left(y\right)=inf\left\{i:{S}_{i,j}\ge y,\left(i,j\right)\in {\mathbb{Z}}_{+}^{2}\right\}$ and
${L}_{J}\left(y\right)=inf\left\{j:{S}_{i,j}\ge y,\left(i,j\right)\in {\mathbb{Z}}_{+}^{2}\right\}$. We also define
${I}_{n}=inf\left\{i:{S}_{i,j}={E}_{{\kappa}_{n}},\left(i,j\right)\in {\mathbb{Z}}_{+}^{2}\right\}$and
${J}_{n}=inf\left\{j:{S}_{i,j}={E}_{{\kappa}_{n}},\left(i,j\right)\in {\mathbb{Z}}_{+}^{2}\right\}$. Let Δ*I*_{i} := *I*_{i}−*I*_{i−1}, the incremental sequence length between (*i*−1)^{th} and *i*^{th} SALEs in sequence **I**, and Δ*J*_{j} := *J*_{j}−*J*_{j−1}, the incremental sequence length between (*j*−1)^{th} and *j*^{th} SALEs in sequence **J**. Last, we define
${\mathbb{E}}^{*}\left[\Delta {I}_{i}\right]:=\mathbb{E}\left[\Delta {I}_{i}{e}^{\lambda {E}_{{\kappa}_{i}}},{\kappa}_{i}<\infty \right]$,
${\mathbb{E}}^{*}\left[\Delta {J}_{j}\right]:=\mathbb{E}\left[\Delta {J}_{j}{e}^{\lambda {E}_{{\kappa}_{j}}},{\kappa}_{j}<\infty \right]$, and
${\mathbb{E}}^{*}\left[\Delta {E}_{i}\right]:=\mathbb{E}\left[\Delta {E}_{i}{e}^{\lambda {E}_{{\kappa}_{i}}},{\kappa}_{i}<\infty \right]$.

The formulas for computing *a*_{I}*a*_{J}*α*_{I}*α*_{J} and *σ* are:

$$\begin{array}{l}{a}_{I}=\underset{i\to \infty}{lim}\frac{{\mathbb{E}}^{*}\left[\Delta {I}_{i}\right]}{{\mathbb{E}}^{*}\left[\Delta {E}_{i}\right]},{a}_{J}=\underset{j\to \infty}{lim}\frac{{\mathbb{E}}^{*}\left[\Delta {J}_{j}\right]}{{\mathbb{E}}^{*}\left[\Delta {E}_{j}\right]},\hfill \\ {\alpha}_{I}=\underset{i\to \infty}{lim}\frac{{var}^{*}\left[\Delta {I}_{i}\right]}{{\mathbb{E}}^{*}\left[\Delta {E}_{i}\right]},{\alpha}_{J}=\underset{j\to \infty}{lim}\frac{{var}^{*}\left[\Delta {I}_{j}\right]}{{\mathbb{E}}^{*}\left[\Delta {E}_{j}\right]},\hfill \\ \sigma =\underset{i\to \infty}{lim}\frac{{cov}^{*}\left[\Delta {I}_{i},\Delta {J}_{i}\right]}{{\mathbb{E}}^{*}\left[\Delta {E}_{i}\right]},\hfill \end{array}$$

(8)

where var^{*} and cov^{*} represent the variance and covariance associated with the probability measure underlying the expectation
${\mathbb{E}}^{*}$. In practice, for computational efficiency, we use importance sampling to estimate the parameters above
[15]. The parameters are estimated separately at each SALE. We then apply asymptotic regression to estimate the values of *a*_{I}*a*_{J}*α*_{I}*α*_{J} and *σ* as *i*→∞ in the equation above
[22].

- Karlin S, Altschul SF. Methods for assessing the statistical significance of molecular sequence features by using general scoring schemes. Proc Natl Acad Sci U S A. 1990;87(6):2264–2268. doi: 10.1073/pnas.87.6.2264. [PubMed] [Cross Ref]
- Galambos J. The asymptotic theory of extreme order statistics. New York: Wiley; 1978.
- Mott R. Maximum-likelihood-estimation of the statistical distribution of Smith-Waterman local sequence similarity scores. Bull Math Biol. 1992;54(1):59–75. [PubMed]
- Waterman MS, Vingron M. Rapid and accurate estimates of statistical significance for sequence data base searches. Proc Natl Acad Sci U S A. 1994;91(11):4625–4628. doi: 10.1073/pnas.91.11.4625. [PubMed] [Cross Ref]
- Altschul SF, Gish W. Local alignment statistics. Methods Enzymol. 1996;266:460–480. [PubMed]
- Bundschuh R. Rapid significance estimation in local sequence alignment with gaps. J Comput Biol. 2002;9(2):243–260. doi: 10.1089/10665270252935449. [PubMed] [Cross Ref]
- Chia N, Bundschuh R. A practical approach to significance assessment in alignment with gaps. J Comput Biol. 2006;13(2):429–441. doi: 10.1089/cmb.2006.13.429. [PubMed] [Cross Ref]
- Newberg LA. Significance of gapped sequence alignments. J Comput Biol. 2008;15(9):1187–1194. doi: 10.1089/cmb.2008.0125. [PMC free article] [PubMed] [Cross Ref]
- Agrawal A, Brendel VP, Huang X. Pairwise statistical significance and empirical determination of effective gap opening penalties for protein local sequence alignment. Int J Computat Biol Drug Des. 2008;1(4):347–367. doi: 10.1504/IJCBDD.2008.022207. [PubMed] [Cross Ref]
- Poleksic A. Island method for estimating the statistical significance of profile-profile alignment scores. BMC Bioinformatics. 2009;10:112. doi: 10.1186/1471-2105-10-112. [PMC free article] [PubMed] [Cross Ref]
- Ortet P, Bastien O. Where does the alignment score distribution shape come from? Evol Bioinformatics. 2010;6:159–187. [PMC free article] [PubMed]
- Agrawal A, Huang X. Pairwise statistical significance of local sequence alignment using sequence-specific and position-specific substitution matrices. IEEE/ACM Trans Comput Biol Bioinformatics. 2011;8(1):194–205. [PubMed]
- Altschul SF. In: Theoretical and computational methods in genome research. Suhai S, editor. New York: Plenum Press; 1997. Evaluating the statistical significance of multiple distinct local alignments; pp. 1–14.
- Altschul SF, Madden TL, Schaffer AA, Zhang J, Zhang Z, Miller W, Lipman DJ. Gapped BLAST and PSI-BLAST: a new generation of protein database search programs. Nucleic Acids Res. 1997;25(17):3389–3402. doi: 10.1093/nar/25.17.3389. [PMC free article] [PubMed] [Cross Ref]
- Park Y, Sheetlin S, Spouge JL. Estimating the gumbel scale parameter for local alignment of random sequences by importance sampling with stopping times. Ann Stat. 2009;37(6A):3697–3714. doi: 10.1214/08-AOS663. [PMC free article] [PubMed] [Cross Ref]
- Asmussen S. Applied probability and queues. New York: Springer; 2003.
- Altschul SF, Bundschuh R, Olsen R, Hwa T. The estimation of statistical parameters for local alignment score distributions. Nucleic Acids Res. 2001;29(2):351–361. doi: 10.1093/nar/29.2.351. [PMC free article] [PubMed] [Cross Ref]
- Hartmann AK. Sampling rare events: statistics of local sequence alignments. Phys Rev E. 2002;65(5) doi: 10.1103/PhysRevE.65.056102. [PubMed] [Cross Ref]
- Chandonia JM, Hon G, Walker NS, Lo Conte L, Koehl P, Levitt M, Brenner SE. The ASTRAL Compendium in 2004. Nucleic Acids Res. 2004;32:D189–D192. doi: 10.1093/nar/gkh034. [PMC free article] [PubMed] [Cross Ref]
- Murzin AG, Brenner SE, Hubbard T, Chothia C. SCOP - a structural classification of proteins database for the investigation of sequences and structures. J Mol Biol. 1995;247(4):536–540. [PubMed]
- Schaffer AA, Aravind L, Madden TL, Shavirin S, Spouge JL, Wolf YI, Koonin EV, Altschul SF. Improving the accuracy of PSI-BLAST protein database searches with composition-based statistics and other refinements. Nucleic Acids Res. 2001;29(14):2994–3005. doi: 10.1093/nar/29.14.2994. [PMC free article] [PubMed] [Cross Ref]
- Sheetlin S, Park Y, Spouge JL. Objective method for estimating asymptotic parameters, with an application to sequence alignment. Phys Rev E. 2011;84(3) doi: 10.1103/PhysRevE.84.031914. [PMC free article] [PubMed] [Cross Ref]

Articles from BMC Research Notes are provided here courtesy of **BioMed Central**

PubMed Central Canada is a service of the Canadian Institutes of Health Research (CIHR) working in partnership with the National Research Council's national science library in cooperation with the National Center for Biotechnology Information at the U.S. National Library of Medicine(NCBI/NLM). It includes content provided to the PubMed Central International archive by participating publishers. |