 Short report
 Open Access
 Published:
Interpol: An R package for preprocessing of protein sequences
BioData Mining volume 4, Article number: 16 (2011)
Abstract
Background
Most machine learning techniques currently applied in the literature need a fixed dimensionality of input data. However, this requirement is frequently violated by real input data, such as DNA and protein sequences, that often differ in length due to insertions and deletions. It is also notable that performance in classification and regression is often improved by numerical encoding of amino acids, compared to the commonly used sparse encoding.
Results
The software "Interpol" encodes amino acid sequences as numerical descriptor vectors using a database of currently 532 descriptors (mainly from AAindex), and normalizes sequences to uniform length with one of five linear or nonlinear interpolation algorithms. Interpol is distributed with open source as platform independent Rpackage. It is typically used for preprocessing of amino acid sequences for classification or regression.
Conclusions
The functionality of Interpol widens the spectrum of machine learning methods that can be applied to biological sequences, and it will in many cases improve their performance in classification and regression.
Findings
Machine learning techniques have been widely applied to biological sequences to gain insights into biological function, for instance Rost and Sander [1], Dubchak et al. [2], Karchin et al. [3] and Nielsen et al. [4]. Nanni and Lumini [5] have found improved performance of classifiers based on numerically encoded amino acid sequences as compared to classifiers based on the typically used standard orthonormal representation, i.e. a vector containing twenty indicator variables (one for each amino acid) for each sequence position, resulting in a matrix containing the amino acid distributions for each position in the input sequence. For numerical encoding, each amino acid (or nucleotide) of a sequence is mapped to a numerical descriptor value, such as hydropathy [6], molecular weight, or isoelectric point.
One major limitation of almost all machine learning algorithms is the fixed input dimension, making these algorithms incapable of handling data which varies in its dimension. This is unsuitable for many biological applications as there are often sequence deletions and insertions.
We have developed a preprocessing approach for machine learning that combines the use of numerical descriptor values with a normalization of sequences to a fixed length by numerical interpolation. This procedure has already been applied to coreceptor usage prediction in HIV1 [7], functional protein classification [8, 9], and HIV1 drug resistance prediction [10] were it led to marked improvements of prediction performance. Although many machine learning algorithms are available as software, no package for the described preprocessing of amino acid sequences is available to date. We have therefore developed Interpol, a flexible and easy to use open source package for the statistical language R http://www.rproject.org/. Currently, Interpol provides encoding of amino acid sequences with 531 different numerical descriptors from the AAindex database [11] and one additional empirical descriptor. Moreover, it allows normalization of encoded sequences to a specific length with five different linear or nonlinear interpolation procedures.
Interpol is included in the Comprehensive R Archive Network (CRAN) and can be directly downloaded and installed by using the following R command:
install.packages("Interpol")
In the following example, we introduce Interpol's two commands AAdescriptor and Interpol applied to a set consisting of 1351 HIV1 V3 loop sequences from Dybowski et al. [7] for the prediction of coreceptor usage (see also Table 1). After loading the set of sequences, the first V3 sequence is encoded using the AAdescriptor command:
library(Interpol)
data(V3) #load V3 data
data.new < AAdescriptor(V3[1]) #numerically encode sequence 1
Optional parameters are the applied descriptor (default descriptor = 151, i.e. the hydropathy scale of Kyte and Doolittle [6]) and an interval normalization (default normalize = 0, i.e. no normalization). The list of available descriptors can be found in data(list).
After encoding the amino acid sequence as numerical vector, it can be normalized to a specific length for subsequent classification. In our example, the V3 sequence lengths vary between 33 and 38 amino acids due to deletions or insertions. The following commands translate the amino acid sequences into numerical sequences using the hydropathy descriptor, and then normalize the sequences to a fixed length of 35:
library (Interpol)
data (V3) #load V3 data
L.norm < 35 #desired length
data.new < matrix(nrow = length(V3),
ncol = L.norm)
for(i in 1:length (V3)){
#AAdescriptor encodes sequences
#Interpol normalizes to length L.norm
data.new[i,] < Interpol (AAdescriptor (V3[i]),
dims = L.norm)
}
Sequence 782 in the V3 dataset has a length of 38 amino acids. In the following example the code for normalization from 38 to 35 amino acids, and for visualization of the interpolation is demonstrated (see Figure 1):
library(Interpol)
data(V3) #load V3 data
sequence < AAdescriptor(data = V3[782], #numerically encoding
descriptor = 151, #hydropathy descriptor
normalize = 2) #interval normalization [0,1]
sequence.35 < Interpol(data = sequence, #normalize sequence
dims = 35, #desired length 35
method="spline") #spline interpolation
plot(sequence, type="l", ylim = c (0,1), #plot sequence 782
ylab="descriptor value",
xlab="sequence position",
lty = 1, lwd = 2)
lines(seq (1,38,(38/35)),sequence.35, #plot normalized sequence
lty = 2, lwd = 2)
axis(3, at = seq (1,38, 38/35), #add axis
labels = 1:35)
The optional parameter method can be one of linear, spline, natural, periodic or fmm (default method = linear). The linear interpolation connects two data points (x _{0}, y _{0}) and (x _{1}, y _{1}) with a straight line. The cubic spline interpolation uses piecewise cubic polynomials instead of a straight line. The spline interpolation of Forsythe [12] builds a cubic spline interpolation with the cubic passing exactly through the four points at each end of a sample. The periodic spline interpolation fulfills periodic boundary conditions, i.e. the spline curve has the same first and second derivative at its endpoints. The natural spline interpolation fulfills the natural boundary conditions.
The command help(package = Interpol) gives an overview of the Interpol package and the included methods and data. Descriptions for the AAdescriptor and the Interpol commands can be obtained by help(AAdescriptor) and help(Interpol), respectively.
In the following examples we demonstrate the use of the Interpol package for the prediction of the coreceptor usage of HIV1 according to Dybowski et al. [7] based on sequences of the V3region of the HIV1 protein gp120. V3 is the main determinant of coreceptor usage, i.e. it determines which of the cellular coreceptors CCR5 or CXCR4 is used by HIV1 for cell entry. Classification of V3sequences with respect to coreceptor usage is important for therapy and prognosis. Since V3 is variable in length, many classification algorithms are not applicable. We therefore first apply AAdescriptor and Interpol to numerically encode V3sequences and to normalize them to a fixed length. We then apply for classification random forests [13] implemented in the randomForest package, and for performance measurement the area under the receiver operating characteristics curve (AUC) implemented in the ROCR package [14] of R according to Dybowski et al. [7]. Note that Interpol is independent of the classification method applied, and could be also used with artificial neural networks (as in Rpackage neuralnet), support vector machines (as in Rpackage kernlab) [15], etc.
library(Interpol)
library(randomForest)
library(ROCR)
data(V3) #load V3 data
desc < 151 #hydropathy descriptor
inter < "linear" #linear interpolation
L.norm < 35 #desired length
classes < c (rep (1,200), rep (0,1151)) #class labels
data.new < matrix (nrow = length (V3),ncol = L.norm)
for(i in 1:length(V3)){
#AAdescriptor encodes sequences
#Interpol normalizes to length L.norm
data.new[i,] < Interpol (AAdescriptor (V3[i],
descriptor = desc),
dims = L.norm,method = inter)
}
rf < randomForest (as.factor (classes)., #build forest ^{~}
data = data.new)
pred < prediction (rf$votes[,2], classes) #prediction object
perf < performance (pred, "auc") #AUC estimation
Using the Interpol package, it is very easy to retrieve and compare the performance of different descriptors, e.g. hydrophobicity and net charge, and different interpolation methods (e.g. linear and spline interpolation), by just changing line desc < 151 to desc < 146 and inter < "linear" to inter < "spline", respectively, in the above code (see also Figure 2). The complete list of descriptors can be found on the help pages (help (Interpol)) and in data(list). Note that the above code somewhat overestimates the true performance as it does not include the leaveonepatientout scheme used by Dybowski et al. [7].
There are several potential limitations of the Interpol method for protein classification. First, normalizing to lengths of less than 50% of the original sequence length will in general lead to loss of information. Thus, we suggest to stretch short sequences to a certain length instead of squeezing longer sequences. However, stretching can also cause problems as the normalized sequence space has a higher dimension and thus classification is more prone to overfitting. A more general limitation of normalization is that in some cases the sequence length itself can carry some information. For instance, classifying sequences of huntingtin protein [16] for induction of Huntington's disease critically relies on the length of a Glutamine repeat, an information that can be partly lost in sequence normalization.
Availability and requirements

Project name: Interpol

Project home page (CRAN): http://cran.rproject.org/web/packages/Interpol/

Operating system (s): Platform independent

Programming language: R (≥ 2.10.0)

License: GPL (≥ 2)

Any restrictions to use by nonacademics: none
References
Rost B, Sander C: Combining evolutionary information and neural networks to predict protein secondary structure. Proteins. 1994, 19: 5572. 10.1002/prot.340190108.
Dubchak I, Muchnik I, Holbrook SR, Kim SH: Prediction of protein folding class using global description of amino acid sequence. Proc Natl Acad Sci. 1995, 92: 87008704. 10.1073/pnas.92.19.8700.
Karchin R, Karplus K, Haussler D: Classifying Gprotein coupled receptors with support vector machines. Bioinformatics. 2002, 18: 147150. 10.1093/bioinformatics/18.1.147.
Nielsen M, Lundegaard C, Worning P, Lauemøller SL, Lamberth K, Buus S, Brunak S, Lund O: Reliable prediction of Tcell epitopes using neural networks with novel sequence representations. Protein Sci. 2003, 12 (5): 10071017. 10.1110/ps.0239403.
Nanni L, Lumini A: A new encoding technique for peptide classification. Expert Systems with Applications. 2011, 38 (4): 31853191. 10.1016/j.eswa.2010.09.005.
Kyte J, Doolittle R: A simple method for displaying the hydropathic character of a protein. J Mol Biol. 1982, 157: 105132. 10.1016/00222836(82)905150.
Dybowski JN, Heider D, Hoffmann D: Prediction of coreceptor usage of HIV1 from genotype. PLoS Comput Biol. 2010, 6 (4): e100074310.1371/journal.pcbi.1000743.
Heider D, Appelmann J, Bayro T, Dreckmann W, Held A, Winkler J, Barnekow A, Borschbach M: A computational approach for the identification of small GTPases based on preprocessed amino acid sequences. Technology in Cancer Research and Treatment. 2009, 8 (5): 333342.
Heider D, Hauke S, Pyka M, Kessler D: Insights into the classification of small GTPases. Advances and Applications in Bioinformatics and Chemistry. 2010, 3: 1524.
Heider D, Verheyen J, Hoffmann D: Machine learning on normalized protein sequences. BMC Research Notes. 2011, 4: 9410.1186/17560500494.
Kawashima S, Pokarowski P, Pokarowska M, Kolinski A, Katayama T, Kanehisa M: AAindex: amino acid index database, progress report 2008. Nucleic Acids Res. 2008, D202D205. 36 Database
Forsythe GE: Computer Methods for Mathematical Computations. 1977, Prentice Hall
Breiman L: Random Forests. Machine Learning. 2001, 45: 532. 10.1023/A:1010933404324.
Sing T, Sander O, Beerenwinkel N, Lengauer T: ROCR: visualizing classifier performance in R. Bioinformatics. 2005, 21 (20): 39403941. 10.1093/bioinformatics/bti623.
Karatzoglou A, Smola A, Hornik K, Zeileis A: kernlab  An S4 Package for Kernel Methods in R. Journal of Statistical Software. 2004, 11 (9): 120.
Walker FO: Huntington's disease. Lancet. 2007, 369 (9557): 218228. 10.1016/S01406736(07)601111.
Acknowledgements
We thank the reviewers for their fruitful suggestions. Funding by Deutsche Forschungsgemeinschaft TRR60/A6 is gratefully acknowledged.
Author information
Authors and Affiliations
Corresponding author
Additional information
Competing interests
The authors declare that they have no competing interests.
Authors' contributions
DH* has implemented and tested the software, and drafted the manuscript. DH has revised 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
Cite this article
Heider, D., Hoffmann, D. Interpol: An R package for preprocessing of protein sequences. BioData Mining 4, 16 (2011). https://doi.org/10.1186/17560381416
Received:
Accepted:
Published:
DOI: https://doi.org/10.1186/17560381416
Keywords
 Spline Interpolation
 Numerical Descriptor
 Coreceptor Usage
 Encode Amino Acid Sequence
 Classification Random Forest