A review of software for microarray genotyping

Transcription

1 SOFTWARE REVIEW A review of software for microarray genotyping Philippe Lamy, 1,2 Jakob Grove 1,3 an Carsten Wiuf 1* 1 Bioinformatics Research Centre, C. F. Mollers Allé 8, Builing 1110, DK-8000 Aarhus C, Denmark 2 Department of Molecular Meicine, Aarhus University Hospital, Skejby, DK-8200 Aarhus N, Denmark 3 Department of Human Genetics, Aarhus University, The Bartholin Builing, Wilhelm Meyers Allé 4, DK-8000, Aarhus C, Denmark *Corresponence to: Tel: þ ; Fax: þ ; wiuf@birc.au.k Date receive (in revise form): 1st March 2011 Abstract The focus of this review is software for the genotyping of microarray single nucleotie polymorphisms, in particular software for Affymetrix an Illumina arrays. Different statistical principles an ieas have been applie to the construction of genotyping algorithms for example, likelihoo versus Bayesian moelling, an whether to genotype one or all arrays at a time. The release of new arrays is generally followe by new, or upate, algorithms. Keywors: SNP array, genotype, calling algorithm, copy number, intensity, software Introuction The use of microarrays an microarray technology in research is now more than 15 years ol an has ha a tremenous impact on many aspects of research. Suenly, it became possible to profile an survey whole genomes an to compare genomes across iniviuals an species to an extent that was harly possible before. The perception of the genome change as genome-wie ata became available to everyone. This review focuses narrowly on software use for genotyping of single nucleotie polymorphisms (SNPs) in connection with SNP microarrays (or arrays for short). There are an estimate ten million or more SNPs in the human genome. 1 For each of these, there are three possible genotypes (assuming iploiy), AA, BB (homozygous) an AB (heterozygous), where A an B enote the two possible alleles. The first commercial SNP array was release in 1996 by Affymetrix (Santa Clara, CA) an targete about 1,500 human SNPs, 2 a tiny fraction of all SNPs. Since then, many ifferent manufacturers have evelope microarrays for genome-wie genotyping, incluing Affymetrix, Agilent (Santa Clara, CA), Illumina (San Diego, CA) an Nimblegen (Maison, WI), with arrays esigne for many ifferent organisms. SNP arrays have foun uses in many research areas an contexts for example, association mapping, 3 linkage isequilibrium mapping, 4 phasing, 5 inference on emography an ancestry, 6 evolution 7 an loss-of-heterozygosity analysis in cancer. 8 Early usage of SNP arrays sought to estimate loss of heterozygosity in cancer by comparing DNA from germline an tumour cells. 9 In aition, SNP arrays have been use to estimate copy numbers in cancers 10 (similar to the use of comparative genomic hybriisation [CGH] arrays) an copy number variants (CNVs) in populations. 11 The newest arrays from Affymetrix an Illumina both contain probes for CNVs an copy number polymorphisms (CNPs). Toay, SNP microarrays are able to genotype more than a million SNPs simultaneously (Table 1). This large number of SNPs poses a number of statistical, as well as computational, problems an has attracte the attention of many statisticians an bioinformaticians. Interestingly, the problems themselves 304 # HENRY STEWART PUBLICATIONS HUMAN GENOMICS. VOL 5. NO MAY 2011

2 A review of software for microarray genotyping SOFTWARE REVIEW Table 1. The arrays that are currently available for the human genome from Affymetrix an Illumina. For Affymetrix, #Arrays reflects the physical number of arrays to use to obtain genotypes for all SNPs. For Illumina, #Samples gives the number of samples that can be run using the same Affymetrix GeneChip Human Mapping 10K 2.0 Array GeneChip Human Mapping 100K Set GeneChip Human Mapping 500K Array Set Genome-Wie Human SNP Array 5.0 Genome-Wie Human SNP Array 6.0 #Arrays #SNPs Software 1 10,204 MPAM 2 116,204 DM 2 500,568 BRLMM 1 500,568 a BRLMM-P 1 906,600 b Birsee Illumina #Samples #Markers Software HumanCytoSNP-12 DNA Analysis Human660W-Qua v1 DNA Analysis HumanOmniExpress Human1M-Duo DNA Analysis HumanOmni1-Qua HumanOmni1S-8 HumanOmni2.5-Qua , , , ,199, ,140, ,200,000 c 4 2,450,000 c a Aitional 420,000 non-polymorphic probes for copy number analysis. b Aitional 946,000 non-polymorphic probes for copy number analysis. c Also inclues probes for CNVs. The BeaStuio an the GenomeStuio applications can hanle all Illumina s arrays. have le to many new evelopments in statistics an have fostere what we might term informatics of large atasets. There are a number of statistical issues that are share between microarrays, irrespective of the platform, chemistry an esign principles. These inclue: (i) (ii) (iii) Normalisation of raw intensities Backgroun correction an outlier etection Genotyping The statistical methos applie at each step are, to some extent, transferable between platforms an array types, in particular the parts relating to (i) an (ii). Normalisation of array intensities is important in orer to make comparisons across arrays. 12,13 Backgroun correction an outlier etection (iniviual ba SNPs, as well as ba arrays) are essential for correct interpretation of the ata 12,13 (ie to reuce the number of false an missing calls). A general review of SNP array platforms an their history an use is given by LaFramboise. 14 Software We focus on the most commonly use platforms, Affymetrix an Illumina, an o not iscuss software for normalisation an backgroun correction. Problem formulation Both platforms represent a SNP by a number of probes (Affymetrix) or beas (Illumina) for each allele. The probes/beas have ifferent affinities, epening on the DNA sequence they target, an thus prouce signals of various strengths (see Figure 1). The newest arrays, Human SNP Array 6.0 from Affymetrix an HumanOmni2.5-Qua from Illumina, use three probes an aroun 20 beas for each allele, respectively. Earlier Affymetrix arrays aitionally use mismatch probes; probes that were esigne to capture nonspecific bining. A first step in many algorithms for genotyping is to summarise the probe intensities for each allele an SNP, an in a secon step to make a call base on the summarise intensities. SNP calling software for Affymetrix SNP arrays Following its release of new SNP arrays (calle GeneChips), Affymetrix has evelope # HENRY STEWART PUBLICATIONS HUMAN GENOMICS. VOL 5. NO MAY

3 SOFTWARE REVIEW Lamy, Grove an Wiuf Figure 1. Normalise an summarise allele intensities from the Illumina array. The intensities are shown in transforme polar coorinates: the theta-coorinate represents the angle from the x-axis (the angle from the x-axis to the vector [A, B] of the two allele intensities), an the R-coorinate represents the copy number (the length of the vector). (A) Intensities for a single nucleotie polymorphism (SNP) from 120 arrays, clearly separating the intensities into three groups (A/A, A/B, B/B). (B) Data from 317,000 SNPs (from the same 120 arrays). This plot clearly inicates that signal strength varies consierably with the SNP, a factor that must be taken into account when genotyping iniviual SNPs an eriving copy numbers. The figure is reprouce with the permission of Gunerson et al. 15 accompanying software that takes into account the properties of the new arrays. The first program, Moifie Partitioning Aroun Meois (MPAM) 16 an Dynamic Moel (DM), 17 were able to genotype one SNP on one chip at a time. The next generation of software, Robust Linear Moel with Mahalanobis Distance Classifier (RLMM), 18 BRLMM 19 (which as a Bayesian step to RLMM), BRLMM-P 20 (which uses perfect match probes only) an Birsee, 21 increase accuracy an performance using a multi-chip approach. One SNP, one chip For the first SNP arrays, Affymetrix esigne software moules (MPAM, DM) to genotype iniviual SNPs, one array at a time. The DM software 17 was introuce with the release of the 100K GeneChip an is base on statistical moelling of quartets. A quartet consists of match an mismatch probes for the two alleles. This software oes not require any normalisation step an oes not summarise the probe intensities. A score is assigne for each quartet an the Wilcoxon signe-ranke test is use to give a call. Multi-chip approach Several groups 3,18,22 28 have esigne SNP calling algorithms using a multi-chip approach; the first algorithm was RLMM. 18 This approach requires pre-processing steps, such as array normalisation, in orer to compare ata across arrays an summation of the probe intensities for each allele. For each SNP, the two allele intensities are then clustere into three clous, representing the ifferent genotypes across many chips (Figure 2). Affymetrix esigne the BRLMM algorithm for the 500K SNP arrays. 19 This algorithm was a significant improvement over the DM algorithm use for the previous arrays. The BRLMM algorithm is an extension of the RLMM software an it uses a Bayesian step to efine cluster centres an variances of SNP intensities. Briefly, after normalisation an allelic summation, genotypes are clustere using a Bayesian prior on cluster centres an variances an a pre-clustering mae by the DM algorithm. The prior is base on a ranom set of SNPs, with a minimum number of iniviuals in each cluster. This allows for a better efinition of the genotype clusters with few ( potentially no) iniviuals. Further, new arrays can be genotype using preefine parameters obtaine from other arrays. For the SNP5.0 GeneChip, Affymetrix esigne a new version of the BRLMM algorithm, name BRLMM-P, as the array oes not have mismatch probes. 20 The DM step of BRLMM is replace 306 # HENRY STEWART PUBLICATIONS HUMAN GENOMICS. VOL 5. NO MAY 2011

4 A review of software for microarray genotyping SOFTWARE REVIEW Other software can be use to genotype SNPs from Affymetrix GeneChips, such as Correcte Robust Moel with Maximum Likelihoo Distance (CRLMM), 23 Genotype calling with Empirical Likelihoo (GEL), 24 SNiPer-High Density (SNiPer-HD) 25, Probe-Level Allele-specific Quantization (PLASQ), 26 MAMS 27 (combines Single-Array Multi-SNP [SAMS] with Multi-Array Single-SNP [MASS]), Chiamo 3 an JAPL. 28 Some work has been one to compare algorithms an, generally, the performance of algorithms are compare with HapMap ata in the original papers. Figure 2. Normalise an summarise allele intensities from the Affymetrix GeneChip array. Each SNP is represente by a pair of intensity values (A, B) for the A an B alleles, respectively (here, on a log-scale). An X chromosome SNP is shown, clearly inicating separation into istinct genotype clusters. The plot also shows that ifferent copy numbers can be istinguishe. Males are haploi for the particular SNP (ie either AY or BY) an show up as homozygous but with reuce allele intensity. Grey: BY; blue: BB; green: AB; re: AA; an pink: AY. by a maximum likelihoo-base ivision into genotype cluster. Further, the prior can be a generic prior common to all SNPs or a SNP-specific prior efine using a set of training ata (such as HapMap ata). For SNP6.0, the Broa Institute, in collaboration with Affymetrix, evelope the Birsuite software. 21,29 The novelty comes from relaxing the assumption that all SNPs are iploi an introucing known CNPs. Birsee is actually compose of four applications: Canary, Birsee, Birseye an Fawkes. Canary can give an estimate of the copy number of known CNPs. Birsee is a genotyping software with use restricte to iploi genomic regions. It is similar to BRLMM. Clusters are preefine using training ata an then further optimise. The Birseye software can etect rare CNVs an genotype SNPs in CNVs. Finally, Fawkes combines the output of the three previous applications to assign a comprehensive genotype (A-null, AA, AB, BB, AAB,...). SNP calling software for Illumina s Illumina has evelope its own software to genotype SNPs on the array. The software is calle GenCall an has not been through the same chain of transformations as the Affymetrix software. The GenCall algorithm was implemente within the BeaStuio application (latest version v3.2.2) 33 but it is now part of the GenomeStuio application (the current version is 1.1.0). It relies on a specific normalisation occurring automatically within the Illumina GenomeStuio software an consists of several steps (incluing outlier removal an backgroun estimation). The normalise intensities are then summarise, such that each SNP is assigne a pair of values corresponing to each allele. This pair represents the allele intensities in polar coorinates; the R-coorinate represents the copy number of the SNP an the theta-coorinate represents the angle from the x-axis (Figure 1). This is a multi-array approach, using information from all arrays simultaneously. The call is mae using a cluster file supplie by Illumina, base on a reference set of samples. There is an option to make the call without using the reference set, instea relying exclusively on the sample arrays, however. This ichotomy is similar to the BRLMM (an subsequent Affymetrix software), where a call can be mae with pre-efine parameters, corresponing to a reference population. Whether one shoul use the reference set for genotype calling epens on the number of sample arrays, the quality of the DNA an the minimal # HENRY STEWART PUBLICATIONS HUMAN GENOMICS. VOL 5. NO MAY

5 SOFTWARE REVIEW Lamy, Grove an Wiuf allele frequency (MAF) of interest, as the size of the reference set etermines the MAF etectable. 34 For SNPs with fewer than three genotype clusters, the locations an variations of the missing genotype clusters are estimate using artificial neural networks. It is also possible manually to change the call of any SNP using Illumina s visualisation tool. For CNV analysis, Illumina has evelope a series of tools which are available as plug-ins to the GenomeStuio genotyping moule. Software for estimation of copy numbers (cnvpartition), etection an annotation of homozygosity in single samples (Homozygosity Detector), etection an annotation of chromosomal aberrations in single samples (ChromoZone) an for calculating a likelihoo score for strength of loss-of-heterozygosity (LOH Score) is available. Other methos have been propose for the arrays. Teo et al. esigne a multi-array genotype calling algorithm (Illuminus) that oes not rely on a reference population. 35 By contrast, Giannoulatou et al. evelope a metho that works entirely within each sample, thereby making the performance inepenent of sample size an of any outsie control samples. 34 Both methos rely on an expectation maximisation (EM) algorithm. The CRLMM algorithm 23 is also available for Illumina ata as a package (GenoSNP) for R/Bioconuctor. 36 Discussion an conclusions The accuracy of genotype calling is usually reporte to be above 99 per cent. This is typically the case when samples an DNA of goo quality are available. Many cancer laboratories are intereste in genotyping SNPs an CNPs, however, as well as estimating copy numbers, from tumour tissue. Here, there are a number of problems that have not yet fully been overcome: tumour tissue typically contains normal cells that are ifficult to remove prior to analysis; also, tumour tissue tens to be heterogeneous, in the sense that ifferent tumour cells have ifferent copy number aberrations. These issues affect the possibility of accurately estimating genotypes an copy numbers, an significantly reuce the accuracy of calling algorithms. We have iscusse software for genotyping; however, much software also has been evelope for further ownstream analysis, to accommoate specific questions an nees. 37,38 Software for normalisation an backgroun correction has likewise receive much attention. These methos are also generally applicable to other types of arrays, an borrowing of ieas between array types is common. The future of SNP arrays, in aition to many other microarray types, such as gene (RNA) expression an microrna expression arrays, is uncertain. For the iniviual laboratory, the common microarray platforms are still more costefficient than the new platforms built on nextgeneration sequencing (NGS) technologies. NGS is alreay ominating research to an extent that few foresaw five years ago, however. In aition, it is possible to have samples sequence through commercial organisations or scientific collaborations. SNP an other arrays are still in use, however. They have transforme the fiel of genomics an sparke an intense interest among the statistics an bioinformatics communities to provie solutions to large-scale ata problems. These solutions are the founation for solving the similar large-scale ata problems encountere with NGS. Acknowlegments The stuy was supporte by grants from the Danish Strategic Research Council ( ) (GEMS consortium), the Danish Cancer Society, the EC project GENICA, the Lunbeck Founation an the John an Birthe Meyer Founation. References 1. Kruglyak, L. an Nickerson, D.A. (2001), Variation is the spice of life, Nat. Genet. Vol. 27, pp Wang, D.G., Fan, J.B., Siao, C.J., Berno, A. et al. (1998), Large-scale ientification, mapping, an genotyping of single-nucleotie polymorphisms in the human genome, Science Vol. 280, pp Wellcome Trust Case Control Consortium (2007), Genome-wie association stuy of 14,000 cases of seven common iseases an 3,000 share controls, Nature Vol. 447, pp Pe er, I., e Bakker, P.I., Maller, J., Yelensky, R. et al. (2006), Evaluating an improving power in whole-genome association stuies using fixe marker sets, Nat. Genet. Vol. 38, pp Niu, T. (2004), Algorithms for inferring haplotypes, Genet. Epiemiol. Vol. 27, pp # HENRY STEWART PUBLICATIONS HUMAN GENOMICS. VOL 5. NO MAY 2011

6 A review of software for microarray genotyping SOFTWARE REVIEW 6. Price, A.L., Butler, J., Patterson, N., Capelli, C. et al. (2008), Discerning the ancestry of European Americans in genetic association stuies, PLoS Genet. Vol. 4, p. e Neafsey, D.E., Schaffner, S.F., Volkman, S.K., Park, D. et al. (2008), Genome-wie SNP genotyping highlights the role of natural selection in Plasmoium falciparum population ivergence, Genome Biol. Vol. 9, p. R Koe, K., Wiuf, C., Christensen, L.L., Wikman, F.P. et al. (2005), High-ensity single nucleotie polymorphism array efines novel stage an location epenent allelic imbalances in human blaer tumors, Cancer Res. Vol. 65, pp Linbla-Toh, K., Tanenbaum, D.M., Daly, M.J., Winchester, E. et al. (2000), Loss-of-heterozygosity analysis of small-cell lung carcinomas using single-nucleotie polymorphism arrays, Nat. Biotechnol. Vol. 18, pp Greenman, C.D., Bignell, G., Butler, A., Ekins, S. et al. (2010), PICNIC: An algorithm to preict absolute allelic copy number variation with microarray cancer ata, Biostatistics Vol. 11, pp Zhang, F., Gu, W., Hurles, M.E. an Lupski, J.R. (2009), Copy number variation in human health, isease, an evolution, Annu. Rev. Genomics Hum. Genet. Vol. 10, pp Bolsta, B.M., Irizarry, R.A., Åstran, M. an Spee, T.P. (2003), A comparison of normalization methos for high ensity oligonucleotie array ata base on variance an bias, Bioinformatics Vol. 19, p Li, C. an Wong, W.H. (2001), Moel-base analysis of oligonucleotie arrays: Moel valiation, esign issues an stanar error application, Genome Biol. Vol. 2, pp. research LaFramboise, T. (2009), Single nucleotie polymorphism arrays: A ecae of biological, computational an technological avances, Nucleic Acis Res. Vol. 37, pp Gunerson, K.L., Kuhn, K.M., Steemers, F.J., Ng, P. et al. (2006), Genotype clustering on HumanHap300 TM, Pharmacogenomics, Vol. 7, pp Liu, W., Di, X., Yang, G., Matsuzaki, H. et al. (2003), Algorithms for large-scale genotyping microarrays, Bioinformatics Vol. 19, pp Di, X., Matsuzaki, H., Webster, T.A., Hubbell, E. et al. (2005), Dynamic moel base algorithms for screening an genotyping over 100K SNPs on oligonucleotie microarrays, Bioinformatics Vol. 21, pp Rabbee, N. an Spee, T.P. (2006), A genotype calling algorithm for Affymetrix SNP arrays, Bioinformatics Vol. 22, pp Affymetrix, Inc. (2006), BRLMM: An improve genotype calling metho for the mapping 500K array set, support/technical/whitepapers/brlmm_whitepaper.pf (last accesse 30th April, 2011). 20. Affymetrix, Inc. (2007), BRLMM-P: A genotype calling metho for the SNP 5.0 array, whitepapers/brlmmp_whitepaper.pf (last accesse 30th April, 2011). 21. Korn, J., Kuruvilla, F.G., McCarroll, S.A., Wysoker, A. et al. (2008), Integrate genotype calling an association analysis of SNPs, common copy number polymorphisms an rare CNVs, Nat. Genet. Vol. 40, pp Lamy, P., Anersen, C.L., Wikman, F.P. an Wiuf, C. (2006), Genotyping an annotation of Affymetrix SNP arrays, Nucleic Acis Res. Vol. 34, p. e Carvalho, B., Spee, T.P. an Irizarry, R.A. (2007), Exploration, normalization, an genotype calls of high-ensity oligonucleotie SNP array ata, Biostatistics Vol. 8, pp Nicolae, D.L., Wu, X., Miyake, K. an Cox, N.J. (2006), GEL: A novel genotype calling algorithm using empirical likelihoo, Bioinformatics Vol. 22, pp Hua, J., Craig, D.W., Brun, M., Webster, J. et al. (2006), SNiPer-HD: Improve genotype calling accuracy by an expectation-maximization algorithm for high-ensity SNP arrays, Bioinformatics Vol. 23, pp LaFramboise, T., Weir, B.A., Zhao, X., Beroukhim, R. et al. (2005), Allele-specific amplification in cancer reveale by SNP array analysis, PLoS Comput. Biol. Vol. 1, p. e Xiao, Y., Segal, M.R., Yang, Y.H. an Yeh, R.-F. (2007), A multi-array multi-snp genotyping algorithm for Affymetrix SNP microarrays, Bioinformatics Vol. 23, pp Plagnol, V., Cooper, J.D., To, J.A. an Clayton, D.G. (2007), A metho to aress ifferential bias in genotyping in large-scale association stuies, PLoS Genet. Vol. 3, p. e McCarroll, S.A., Kuruvilla, F.G., Korn, J.M., Cawley, S. et al. (2008), Integrate etection an population-genetic analysis of SNPs an copy number variation, Nat. Genet. Vol. 40, pp Lin, S., Carvalho, B., Cutler, D., Arking, D. et al. (2008), Valiation an extension of an empirical Bayes metho for SNP calling on Affymetrix microarrays, Genome Biol. Vol. 9, pp Kim, J.-H., Jung, S.-H., Hu, H.-J., Yim, S.-H. et al. (2010), Comparison of the Affymetrix SNP Array 5.0 an oligoarray platforms for efining CNV, Genomics Informatics Vol. 8, pp Vens, M., Schillert, A., König, I.R. an Ziegler, A. (2009), Look who is calling: A comparison of genotype calling algorithms, BMC Proc. Vol. 3, p. S Steemers, F.J. an Gunerson, K.L. (2007), Whole genome genotyping technologies on the BeaArray platform, Biotechnol. J. Vol. 2, pp Giannoulatou, E., Yau, C., Colella, S., Ragoussis, J. et al. (2008), GenoSNP: A variational Bayes within-sample SNP genotyping algorithm that oes not require a reference population, Bioinformatics Vol. 24, pp Teo, Y.Y., Inouye, M., Small, K.S., Gwilliam, R. et al. (2007), A genotype calling algorithm for the Illumina BeaArray platform, Bioinformatics Vol. 23, pp Ritchie, M.E., Carvalho, B.S., Hetrick, K.N. an Tavaré, S. (2009), R/Bioconuctor software for Illumina s Infinium whole-genome genotyping s, Bioinformatics Vol. 25, pp Aroma.affymetrix, web/software?version=5&pli=1 (last accesse 30th April, 2011). 38. Cheng Li Lab, (last accesse 30th April, 2011). # HENRY STEWART PUBLICATIONS HUMAN GENOMICS. VOL 5. NO MAY