MarkerMatch: A Proximity-Based Probe-Matching Algorithm for Joint Analysis of Copy-Number Variants from Different Genotyping Arrays

MarkerMatch: A Proximity-Based Probe-Matching Algorithm for Joint Analysis of Copy-Number Variants from Different Genotyping Arrays | bioRxiv /* */ /* */ <!-- <!-- /*! * yepnope1.5.4 * (c) WTFPL, GPLv2 */ (function(a,b,c){function d(a){return"[object Function]"==o.call(a)}function e(a){return"string"==typeof a}function f(){}function g(a){return!a||"loaded"==a||"complete"==a||"uninitialized"==a}function h(){var a=p.shift();q=1,a?a.t?m(function(){("c"==a.t?B.injectCss:B.injectJs)(a.s,0,a.a,a.x,a.e,1)},0):(a(),h()):q=0}function i(a,c,d,e,f,i,j){function k(b){if(!o&&g(l.readyState)&&(u.r=o=1,!q&&h(),l.onload=l.onreadystatechange=null,b)){"img"!=a&&m(function(){t.removeChild(l)},50);for(var d in y[c])y[c].hasOwnProperty(d)&&y[c][d].onload()}}var j=j||B.errorTimeout,l=b.createElement(a),o=0,r=0,u={t:d,s:c,e:f,a:i,x:j};1===y[c]&&(r=1,y[c]=[]),"object"==a?l.data=c:(l.src=c,l.type=a),l.width=l.height="0",l.onerror=l.onload=l.onreadystatechange=function(){k.call(this,r)},p.splice(e,0,u),"img"!=a&&(r||2===y[c]?(t.insertBefore(l,s?null:n),m(k,j)):y[c].push(l))}function j(a,b,c,d,f){return q=0,b=b||"j",e(a)?i("c"==b?v:u,a,b,this.i++,c,d,f):(p.splice(this.i++,0,a),1==p.length&&h()),this}function k(){var a=B;return a.loader={load:j,i:0},a}var l=b.documentElement,m=a.setTimeout,n=b.getElementsByTagName("script")[0],o={}.toString,p=[],q=0,r="MozAppearance"in l.style,s=r&&!!b.createRange().compareNode,t=s?l:n.parentNode,l=a.opera&&"[object Opera]"==o.call(a.opera),l=!!b.attachEvent&&!l,u=r?"object":l?"script":"img",v=l?"script":u,w=Array.isArray||function(a){return"[object Array]"==o.call(a)},x=[],y={},z={timeout:function(a,b){return b.length&&(a.timeout=b[0]),a}},A,B;B=function(a){function b(a){var a=a.split("!"),b=x.length,c=a.pop(),d=a.length,c={url:c,origUrl:c,prefixes:a},e,f,g;for(f=0;f<d;f++)g=a[f].split("="),(e=z[g.shift()])&&(c=e(c,g));for(f=0;f<b;f++)c=x[f](c);return c}function g(a,e,f,g,h){var i=b(a),j=i.autoCallback;i.url.split(".").pop().split("?").shift(),i.bypass||(e&&(e=d(e)?e:e[a]||e[g]||e[a.split("/").pop().split("?")[0]]),i.instead?i.instead(a,e,f,g,h):(y[i.url]?i.noexec=!0:y[i.url]=1,f.load(i.url,i.forceCSS||!i.forceJS&&"css"==i.url.split(".").pop().split("?").shift()?"c":c,i.noexec,i.attrs,i.timeout),(d(e)||d(j))&&f.load(function(){k(),e&&e(i.origUrl,h,g),j&&j(i.origUrl,h,g),y[i.url]=2})))}function h(a,b){function c(a,c){if(a){if(e(a))c||(j=function(){var a=[].slice.call(arguments);k.apply(this,a),l()}),g(a,j,b,0,h);else if(Object(a)===a)for(n in m=function(){var b=0,c;for(c in a)a.hasOwnProperty(c)&&b++;return b}(),a)a.hasOwnProperty(n)&&(!c&&!--m&&(d(j)?j=function(){var a=[].slice.call(arguments);k.apply(this,a),l()}:j[n]=function(a){return function(){var b=[].slice.call(arguments);a&&a.apply(this,b),l()}}(k[n])),g(a[n],j,b,n,h))}else!c&&l()}var h=!!a.test,i=a.load||a.both,j=a.callback||f,k=j,l=a.complete||f,m,n;c(h?a.yep:a.nope,!!i),i&&c(i)}var i,j,l=this.yepnope.loader;if(e(a))g(a,0,l,0);else if(w(a))for(i=0;i (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0];var j=d.createElement(s);var dl=l!='dataLayer'?'&l='+l:'';j.src='//www.googletagmanager.com/gtm.js?id='+i+dl;j.type='text/javascript';j.async=true;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-M677548'); Skip to main content Home About Submit ALERTS / RSS Search for this keyword Advanced Search New Results MarkerMatch: A Proximity-Based Probe-Matching Algorithm for Joint Analysis of Copy-Number Variants from Different Genotyping Arrays View ORCID Profile Franjo Ivankovic , View ORCID Profile Dongmei Yu , View ORCID Profile James Shen , View ORCID Profile Lingyu Zhan , View ORCID Profile Maria Niarchou , View ORCID Profile Ariadne Kaylor , View ORCID Profile Laura Domènech , View ORCID Profile Tyne W Miller-Fleming , View ORCID Profile Luz M Porras , View ORCID Profile Paola Giusti-Rodríguez , View ORCID Profile Roel A Ophoff , View ORCID Profile Jeremiah M Scharf , View ORCID Profile Carol A Mathews doi: https://doi.org/10.1101/2025.06.30.662249 Franjo Ivankovic 1 Center for OCD, Anxiety, and Related Disorders, Department of Psychiatry, McKnight Brain Institute, College of Medicine, University of Florida , Gainesville, FL 32610 2 University of Florida Genetics Institute, Genetics and Genomics Graduate Program, College of Medicine, University of Florida , Gainesville, FL 32610 3 Analytic and Translational Genetics Unit, Massachusetts General Hospital , Boston, MA 02114 4 Stanley Center for Psychiatric Research, Broad Institute of MIT and Harvard , Cambridge, MA 02142 5 Molecular and Population Genetics Program, Program in Brain Health, Broad Institute of MIT and Harvard , Cambridge, MA 02142 Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Franjo Ivankovic For correspondence: franjo{at}ivankovic.us Dongmei Yu 4 Stanley Center for Psychiatric Research, Broad Institute of MIT and Harvard , Cambridge, MA 02142 6 Psychiatric and Neurodevelopemental Genetics Unit, Center for Genomic Medicine, Departments of Psychiatry and Neurology, Massachusetts General Hospital , Boston, MA 02114 Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Dongmei Yu James Shen 1 Center for OCD, Anxiety, and Related Disorders, Department of Psychiatry, McKnight Brain Institute, College of Medicine, University of Florida , Gainesville, FL 32610 Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for James Shen Lingyu Zhan 7 Center for Neurobehavioral Genetics, Semel Institute for Neuroscience and Human Behavior, University of California, Los Angeles , Los Angeles, CA 90024 8 The Collaboratory, Institute for Quantitative and Computational Biosciences, University of California , Los Angeles, Los Angeles, CA, 90095 Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Lingyu Zhan Maria Niarchou 9 Department of Medicine, Division of Genetic Medicine, Vanderbilt University Medical Center , Nashville, TN 32703 Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Maria Niarchou Ariadne Kaylor 4 Stanley Center for Psychiatric Research, Broad Institute of MIT and Harvard , Cambridge, MA 02142 Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Ariadne Kaylor Laura Domènech 4 Stanley Center for Psychiatric Research, Broad Institute of MIT and Harvard , Cambridge, MA 02142 10 Broad Institute of MIT and Harvard, Cambridge, MA 02142; Department of Human Genetics, David Geffen School of Medicine, University of California Los Angeles , Los Angeles, CA 90095 Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Laura Domènech Tyne W Miller-Fleming 9 Department of Medicine, Division of Genetic Medicine, Vanderbilt University Medical Center , Nashville, TN 32703 Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Tyne W Miller-Fleming Luz M Porras 1 Center for OCD, Anxiety, and Related Disorders, Department of Psychiatry, McKnight Brain Institute, College of Medicine, University of Florida , Gainesville, FL 32610 Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Luz M Porras Paola Giusti-Rodríguez 1 Center for OCD, Anxiety, and Related Disorders, Department of Psychiatry, McKnight Brain Institute, College of Medicine, University of Florida , Gainesville, FL 32610 Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Paola Giusti-Rodríguez Roel A Ophoff 7 Center for Neurobehavioral Genetics, Semel Institute for Neuroscience and Human Behavior, University of California, Los Angeles , Los Angeles, CA 90024 Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Roel A Ophoff Jeremiah M Scharf 4 Stanley Center for Psychiatric Research, Broad Institute of MIT and Harvard , Cambridge, MA 02142 6 Psychiatric and Neurodevelopemental Genetics Unit, Center for Genomic Medicine, Departments of Psychiatry and Neurology, Massachusetts General Hospital , Boston, MA 02114 12 Division of Movement Disorders, Department of Neurology, Massachusetts General Hospital, Harvard Medical School , Boston, MA 02114 13 Division of Cognitive and Behavioral Neurology, Brigham and Women’s Hospital, Harvard Medical School , Boston, MA 02115 Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Jeremiah M Scharf Carol A Mathews 1 Center for OCD, Anxiety, and Related Disorders, Department of Psychiatry, McKnight Brain Institute, College of Medicine, University of Florida , Gainesville, FL 32610 2 University of Florida Genetics Institute, Genetics and Genomics Graduate Program, College of Medicine, University of Florida , Gainesville, FL 32610 14 Evelyn F. and William L. McKnight Brain Institute, University of Florida , Gainesville, FL 32610 Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Carol A Mathews Abstract Full Text Info/History Metrics Supplementary material Data/Code Preview PDF Abstract Motivation Copy-number variants (CNVs) are a form of genetic structural variation with increasing importance in complex human disorders. Both DNA sequencing and microarray data can be used to call CNVs, which can be used in association tests, such as association between CNV number and disease status. Unlike genotypes, CNV detection in microarrays requires the use of observed intensity signals at each probe, which limits the imputability for analyses that span multiple array types. Thus far, a consensus set of probes (the intersection encompassing the probes that occur in common on all arrays) has been used to circumvent the problem of differing array-specific sensitivities. This has, however, led to excessive reduction in overall sensitivity of CNV calls as arrays can have an undesirably low overlap of probe sets. To overcome this limitation, we developed MarkerMatch, a proximity-based algorithm that matches probes across different genotyping microarrays to maximize the number of probes considered in the CNV calling algorithm, thereby increasing the resolution and sensitivity while preserving precision. Results By analyzing CNV calls from 4,906 individuals genotyped across three different arrays (Global Screening Array, Omni2.5 array, and Omni Express Exome array), we show that the MarkerMatch approach improves sensitivity by increasing the density of probes available for CNV calling while maintaining precision or improving it relative to the current practice (e.g., use of consensus probes only). We further demonstrate that MarkerMatch exceeds the output from current practice in terms of F1 score, Fowlkes-Mallows index, and Jaccard index. We also optimize MarkerMatch parameters, D MAX and Method , and find an optimal D MAX setting at 10kb, with no clear optimal candidate based on Method , indicating that parameters for this metric should be determined on a use case basis. Introduction Copy-number variants (CNVs) are a form of structural variant involving unbalanced rearrangements leading to increased (duplication) or decreased (deletion) DNA content ( Zarrei et al . 2015 ). CNVs have been studied in the context of various complex human disorders to better understand their underlying pathobiology ( Lionel et al . 2011 ; Olson et al . 2014 ; Huang et al . 2017 ; Marshall et al . 2017 : 17; Wang et al . 2018 ; Nakatochi, Kushima and Ozaki 2021 ; Fu et al . 2022 ). Microarrays are one of the most commonly used technologies to assay CNVs due to their relatively low cost and widespread use in genome-wide association studies and biobanks such as All of Us and the UK Biobank. ( Bycroft et al . 2018 ; The All of Us Research Program Genomics Investigators et al . 2024). While microarrays were not designed for the specific purpose of assaying CNVs, several algorithms have been developed to accurately assess CNV events using microarray data. Some software, e.g., PennCNV, QuantiSNP, Birdseye, and GenoCN, employ Bayesian approaches, such as Hidden Markov Models (HMM), to call CNVs using the intensity data from genotyping microarrays ( Colella et al . 2007 ; Wang et al . 2007 ; McCarroll et al . 2008 ; Sun et al . 2009 ). Other approaches like cnvPartition or iPattern employ recursive partitioning and/or clustering to determine copy-number states ( Pinto et al . 2011 ; Illumina 2017). More recent methods have focused on combining or building on existing approaches to fine-tune CNV calling performance ( Zhang et al . 2019 ; Lavrichenko et al . 2021 ). Genotyping microarrays vary in the density and selection of probes, with some arrays designed to capture variation specific to populations, diseases, genomic regions, etc ( Ehli et al . 2017 ). This variability presents challenges to meta-analysis efforts, which are crucial for aggregating sufficiently powered datasets to detect genetic associations in complex traits ( Zeggini and Ioannidis 2009 ). By leveraging the effects of linkage disequilibrium, accurate imputation of non-genotyped markers is possible to enable joint and meta-analysis across different genotyping arrays ( Scheet and Stephens 2006 ; Marchini and Howie 2010 ). However, CNV detection algorithms rely on direct probe intensity measures that may vary across arrays, and which cannot be imputed for CNV analysis. Probe density and distribution vary across different array products, which may lead to array-type biases in CNV call sensitivity and specificity ( Wang et al . 2007 ). Traditionally, to avoid such biases, researchers have taken manifest intersections and focused only on markers that were genotyped on all SNP arrays for CNV detection and analyses ( Huang et al . 2017 ). However, this approach works only if the genotyping microarrays considered are highly similar; if the overlap between probes across the genotyping microarrays is low, then the overall resolution and sensitivity of CNV detection and analyses can be severely impacted. To overcome this limitation, we developed MarkerMatch. MarkerMatch is a proximity-based algorithm that matches probes across different genotyping microarrays within specified genomic distance (a distance between a probe on reference array and those on the matched array) to maximize the number of probes considered in the CNV calling algorithm, thereby increasing the resolution and sensitivity of subsequent genome-wide CNV association analyses without affecting their specificity. MarkerMatch returns a list of probes for each of the matched arrays to be used in CNV calling. We tested MarkerMatch in two independent experiments: a within-array experiment to test the effects of probe downsampling on CNV calling, and a cross-array experiment to test the reliability of CNV calling in samples genotyped on multiple arrays. Materials and Methods Samples and Software Samples from two independent cohorts (totalling 4,906 individuals) and three different Illumina array products (Global Screening Array, GSA; Omni2.5 array, OMNI; and Omni Express Exome array, OEE) were used to test and validate MarkerMatch algorithms. A summary of the cohorts, specific genotyping platforms, sample sizes, and probe density information is shown in Table 1 . Briefly, we used existing genomic data from the Simons Simplex Collection (SSC) and Tourette Association of America International Consortium for Genetics (TAAICG), detailed descriptions of which are provided in the Supplementary Information ( Fischbach and Lord 2010 ; Scharf et al . 2013 ; Sanders et al . 2015 ; Huang et al . 2017 ). Table 2 summarizes software used in this study. View this table: View inline View popup Download powerpoint Table 1. Samples used in this study. View this table: View inline View popup Download powerpoint Table 2. Software used in this study. Development of MarkerMatch MarkerMatch follows a simple loop and match algorithm ( Figure 1A ) to identify the best-matching probes between two manifests at a time. MarkerMatch takes in two annotated manifests, the smaller of which is considered the reference manifest and the larger the matching manifest. Both reference and matching manifests must contain the following information: (1) probe name, (2) chromosome, (3) genomic position, (4) Download figure Open in new tab Figure 1. Panel A . Diagram depicting the MarkerMatch algorithm. MarkerMatch follows a step-by-step process to identify the best match for each probe across the two selected manifests, while ensuring no duplicates. Specifically, in the first step (exact matching) MarkerMatch will take an intersection of probes from two manifests and keep them in the output . The second step ( Method matching): MarkerMatch will take a probe from the reference manifest and match it with all the remaining probes (those not used in the first step) in the matching manifest within the specified D MAX distance. A probe from the matched manifest that has the smallest difference in the selected Method from the selected probe from the reference manifest will be retained. Once a probe from the matching manifest is paired with a reference probe, it is removed from the matching manifest. This prevents it from being matched again, avoiding repetitive matching of identical probes from the matched manifest. This process will continue until all reference probes have been considered. Panel B . Graphical representation of experimental setup. Blue boxes represent unprocessed array data, with red borders representing reference manifests and green borders representing matching manifests. Yellow boxes represent the MarkerMatch algorithm for WAE (for all Methods and 10bp < D MAX < 5Mb) and CAE (for all Methods and D MAX = 10kb). Green boxes represent output manifests of the MarkerMatch algorithm (-MAT suffix indicates output manifests from MarkerMatch ). Red boxes represent exact match manifests as currently used in CNV association analyses (intersections, also consensus manifests and-EM suffix). In WAE, we compared matched OMNI callsets to full OMNI as a truth set (also Full Set ). In CAE, we compared matched GSA callsets to full OEE as a truth set, as well as matched OEE callsets to full GSA as a truthset. OMNI: Omni2.5 array, GSA: Global Screening Array, OEE: Omni Express Exome array. These processes have been repeated for each iteration of Method and D MAX combination. B-allele frequency (BAF), (5) mean of the log-R ratio (LRR mean), and (6) standard deviation of LRR (LRR sd). In addition to the two manifests, the MarkerMatch function requires two pre-specified parameters: D MAX (which determines the maximum allowable distance from which a marker can be selected) and Method (which determines what metric should be prioritized for matching), and returns a 1:1 set of matched probes from the two manifests. There are 4 options for the Method parameter in the MarkerMatch function that can be used to select probes: position , BAF , LRR mean , and LRR sd . MarkerMatch follows a step-by-step process to identify the best match for each probe across the two selected manifests, while ensuring no duplicates. Specifically, in the first step (exact matching) MarkerMatch will take an intersection of probes from two manifests and keep them. In the second step (nearby matching), MarkerMatch will take a probe from the reference manifest and match it with all the remaining probes (those not used in the first step) in the matching manifest within the specified D MAX distance. A probe from this set that has the smallest difference in the selected Method from the identified probe in the reference manifest will be selected and saved into the output manifest . Once a probe from the matching manifest is paired with a reference probe, it is removed from the matching manifest to avoid it being matched again. This process will continue until all reference probes have been considered. MarkerMatch was written as an R script dependent only on tidyverse packages ( Wickham et al . 2019 ) and is easy and flexible to implement. An alternative implementation in Python has also been written. CNV Calling Briefly, the Illumina GenomeStudio final reports were exported from for each array and passed into PennCNV to call CNVs ( Wang et al . 2007 ; Illumina 2020). Data preprocessing, array clustering, genotyping quality control, CNV calling and quality control are described in detail in the Supplementary Information. For the Within-Array Experiment (WAE) utilizing SSC data, we performed CNV calling for the GSA-matched OMNI manifest at variable MarkerMatch matching metrics (position, LRR mean, LRR standard deviation, and BAF) and variable maximum allowable distances (10bp, 50bp, 100bp, 500bp, 1kb, 5kb, 10kb, 50kb, 100kb, 500kb, 1Mb, and 5Mb), as well as for the full OMNI manifest (full set) and for the intersection of the OMNI manifest with the GSA manifest (exact match). We additionally modeled validation metric performance of MarkerMatch callsets relative to the Full Set to determine the optimal D MAX parameter setting. For the Cross-Array Experiment (CAE) utilizing TAAICG data, we performed CNV calling for the full OEE manifest (full OEE set), the full GSA manifest (full GSA set), an intersection of the OEE manifest with the GSA manifest (exact match), as well as the GSA-matched OEE manifest at a fixed maximum allowable distance of 10kb and variable MarkerMatch matching metrics (position, LRR mean, LRR standard deviation, and BAF). Validation We performed two independent validations of the MarkerMatch: the Within-Array Experiment (WAE) examined MarkerMatch’s performance in probe reductions within the same array (OMNI data from SSC), and the Cross-Array Experiment (CAE), which examined MarkerMatch’s performance across arrays using the OEE and GSA data from TAAICG. A detailed explanation of these two experiments is provided in the Supplementary Information and the graphical representation is shown in Figure 1B . For each experiment, we derived a partial confusion matrix including true positive, false positive, and false negative counts. Truth sets were full set OMNI, OEE, and GSA CNV callsets. True negative counts were impossible to determine as we do not know the true copy-number states for the examined genomes. Based on the partial confusion matrix, we derived the following metrics: true positive rate (sensitivity, recall), false negative rate (FNR), positive predictive value (PPV, precision), false discovery rate (FDR), F1 score (F1; harmonic mean of precision and recall), Fowlkes–Mallows index (FMI; geometric mean of precision and recall); and Jaccard index (JI; ratio of the intersection to the union of the two sets). These data were used to visually and quantitatively inspect performance of specific D MAX and Method parameter configurations in MarkerMatch callsets, and to inform decisions for optimal parameter selection. The full methodology is available in the Supplementary Information. Results Implementation of MarkerMatch The MarkerMatch algorithm runs about 4 times faster in Python than R ( Figure 2 ) across all Methods , with an average R run time of 2.19 minutes and an average Python run time of 0.54 minutes on chromosome 22. The total genome-wide run for MarkerMatch with parameters D MAX = 10kb and Method = Distance was 20.52 hours in R and 7.22 hours in Python. Download figure Open in new tab Figure 2. Execution time curves for chromosome 22, shown as function of time (in minutes) on the y-axis, and log 10 of maximum matching distance D MAX (in bp) on x-axis. Solid lines represent Python, while dashed lines represent R execution times. Method is shown in colors (BAF in yellow, LRR mean in purple, LRR sd in green, and Distance in brown). Analysis of array coverage shows successful recovery of GSA coverage when matched with both the OMNI and OEE arrays ( Figure 3A-D , Table 3 , Supplemental Table S1B). For the OMNI array, coverage plateaued at a D MAX value of 10kb, with 28% coverage of the OMNI array ( Figure 3A ) and 98% coverage of the GSA array ( Figure 3B ). In contrast, the exact match approach resulted in retention of 5% of markers from the OMNI array ( Figure 3A ) and 19% of markers from the GSA array ( Figure 3B ). Download figure Open in new tab Figure 3. Figure showing coverage of arrays in Within-Array Experiment (WAE; A: OMNI, B: GSA) and Cross-Array Experiment (CAE; C: OEE, D: GSA). In all graphs, y-axis is showing the coverage rates and the x-axis is showing maximum allowable distances in base-pairs, log 10 (D MAX ). Note: Lines for all matching Methods in panels A and B are overlapped. Points on panels C and D are horizontally jittered for visibility, but log 10 (D MAX ) is 4 for all matching Methods. OMNI: Omni2.5 array, GSA: Global Screening Array, OEE: Omni Express Exome array. View this table: View inline View popup Download powerpoint Table 3. Summaries of MarkerMatch outcomes at D MAX = 10kb across all Method parameters (BAF, LRR mean, LRR sd, and Distance), as well as Full Set and Exact Match reference comparisons. Coverage is displayed in rate; gaps and LRR sds are displayed in log 10 [median] (log 10 [IQR]); BAFs and LRR means are displayed in median (IQR). Within-array experiment outcomes are detailed in OMNI array, OMNI matched to GSA and GSA array, OMNI matched to GSA segments. Cross-array experiment outcomes are detailed in OEE array, OEE matched to GSA and GSA array, OEE matched to GSA segments. For the OEE array, coverage at D MAX = 10kb was 63% of the OEE array ( Figure 3C ) and 88% of the GSA array ( Figure 3D ). The exact match approach resulted in the retention of 14% markers from the OEE array ( Figure 3C ) and 20% markers from the GSA array ( Figure 3D ). Inter-marker gaps (i.e., gaps between SNPs used by PennCNV to make CNV calls) were also reduced in size on both the OMNI (median 2.2kb at D MAX = 10kb) and OEE (median 2.2kb at D MAX = 10kb) arrays matched to the GSA array, relative to their Exact Match counterparts with a median of about 11kb (Supplemental Figure S1A-C, Table 3 , Supplemental Table S1C). BAF distributions of Full Set arrays were better approximated by MarkerMatched than Exact Match configurations (Supplemental Figure S2A-C, Table 3 , Supplemental Table S1D). Notably, the median BAF differed for each Method , with the median BAF being 0.04 for Method = BAF, 0.08 for Method = LRR mean, 0.11 for Method = LRR sd, and 0.11 for Method = Distance arrays, whereas the median BAF values for Exact Match and Full Set were 0.26 and 0.07, respectively. Conversely, LRR sds and LRR means showed relatively little variability between various Method and D MAX configurations of MarkerMatch, as well as between MarkerMatch configurations and Full Set and Exact Match , with median LRR sd values around 1.26 and LRR mean values around-0.002 (Supplemental Figures S3A-C, S5A-C, S6A-C, Table 3 , Supplemental Tables S1E, S1F). Within-Array Experiment (WAE) Within-Array Experiment (WAE) results are summarized in Table 4 , and in Supplemental Tables S1G-I. Briefly, the Full Set callset resulted in the most CNV calls (low-stringency QC = 288,602; medium-stringency QC = 56,017) and Exact Match resulted in the fewest (low-stringency QC = 11,173; medium-stringency QC = 6,776). MarkerMatched callsets counted 3-8 times more CNV calls relative to Exact Match across all four Methods at D MAX = 10kb (low-stringency QC range 77,132 - 84,140; medium-stringency QC range 31,430 - 34,272). View this table: View inline View popup Table 4. Summaries of MarkerMatch CNV callsets at D MAX = 10kb across all Method parameters (BAF, LRR mean, LRR sd, and Distance), as well as Full Set and Exact Match reference comparisons for OMNI array. Full tables including other D MAX parameter values and specific size bins, as well as other statistics like medians and IQRs, are available in the Supplemental Tables S1G-I. Note: callset PPVs are based on comparisons to OMNI array Full Set callset. In the QC column, low stands for low-stringency QC and med stands for medium-stringency QC. WAE per-sample CNV calls, summarized in Table 4 and Supplemental Table S1G, were highest for the Full Set callset (low-stringency QC = 68.1; medium-stringency QC = 13.2), lowest for Exact Match (low-stringency QC = 2.6, medium-stringency QC = 1.6), and about 3-7 times the Exact Match in MarkerMatched callsets across all four Methods at D MAX = 10kb (low-stringency QC range 18.2 - 19.9; medium-stringency QC range 7.4 - 8.1). The average CNV sizes (in bp) identified in WAE were smaller on denser MarkerMatched configurations ( Table 4 , Supplemental Table S1G). Full Set averages were the smallest (low-stringency QC 25,405.7 bp; medium-stringency QC 95,946.6 bp), Exact Match were the largest (low-stringency QC 116,693.4bp; medium-stringency QC 148,788.1bp), and the MarkerMatched callsets were intermediate (low-stringency QC range 55,342.0bp - 54,799.7bp; medium-stringency QC 102,544.3bp - 105,736.8bp). Average confidence scores in WAE were larger on denser configurations ( Table 4 , Supplemental Table S1G), ranging from 122.6 on medium-stringency QC Full Set to 31.7 on low-stringency QC Exact Match . As with other metrics, CNV confidence scores were intermediate for MarkerMatch callsets (low-stringency QC range 32.3 - 35.1; medium-stringency QC range 50.6 - 57.1). Overall number of samples with at least one CNV call were consistent after low-stringency QC across all approaches (N = 4,239), although they did vary somewhat after medium-stringency QC (N range 3,449 - 3,917) as shown in Table 4 and Supplemental Table S1H. Sample-specific metrics did not vary substantially across low-stringency QC/medium-stringency QC or Full Set / Exact Match /MarkerMatched configurations, with average LRR means at-0.004, LRR sds at 0.114-0.121, BAF means at 0.503, BAF sds at 0.002, BAF drifts at 0.000, and WFs at 0.000-0.001 ( Table 4 , Supplemental Table S1H). Genome-Wide Validation WAE genome-wide validation resulted in higher PPVs for MarkerMatch callsets matched at D MAX = 10kb for all Method parameters (ranges 0.896 - 0.907 and 0.943 - 0.958 for low-and medium-stringency QC, respectively) compared to Exact Match (0.824 and 0.900 for low-and medium-stringency QC, respectively), and for low-stringency QC and medium-stringency QC callsets, respectively ( Figure 4 , Table 4 , Supplemental Table S1I). Download figure Open in new tab Figure 4. Within-Array Experiment (WAE) positive predictive value (PPV) plots for both deletions and duplications. Panels A-F show PPV metrics stratified by CNV size. Dashed line represents low-stringency QC, solid line represents medium-stringency QC. Other metrics performed similarly, with larger CNVs having consistently better performance than smaller CNVs (Supplemental Figures S7-12, Supplemental Table S1I). Additionally, duplications had a slightly better performance than deletions (Supplemental Figure S12-25, Supplemental Table S1I). The genome-wide PPV plots ( Figure 4 ) indicated that, in the majority of D MAX and Method parameter configurations, MarkerMatch somewhat or substantially outperformed the Exact Match approach. Regional Validation Filtering on the medium-stringency QC Exact Match callset resulted in region-specific PPVs of 0.99 (telomeric), 0.90 (centromeric), and 0.96 (segmental duplications), whereas the genome-wide, unfiltered callset had a PPV of 0.90. Performance of MarkerMatch callsets at D MAX = 10kb was consistent across various Method parameters in the medium-stringency QC callset, with region-specific average PPVs of 0.98 (telomeric), 0.93 (centromeric), and 0.87 (segmental duplications). Detailed reports for other D MAX parameter settings and performance metrics are available in the Supplemental Table S1J. These data are graphically shown for all validation metrics in the Supplemental Figures S26-32. Selection of D MAX Visual inspection of plotted validation metrics after loess -smoothing and Full Set scaling (see the supplement equation Eq. 8) indicates that, for the majority of metrics (across sensitivity, PPV, F1, FMI, and JI), regardless of CNV size (all, CNV < 100kb, 100kb < CNV < 500kb, or 500kb < CNV < 1Mb), CNV type (all, deletions, or duplications), or Method parameter (BAF, LRR mean, LRR sd, or Distance), the peak, plateau, and/or inflection point occurred in the range of D MAX between 10kb and 100kb (Supplemental Figures S32-37). Further inspection indicated that D MAX = 10kb was the optimal maximum allowable distance to match within, and was thus chosen as the D MAX setting for the Cross-Array Experiment (CAE). Selection of Method Visual inspection of the plotted validation metrics after Full Set scaling (see the supplement equation Eq. 8) indicated that, for the majority of metrics (across sensitivity, PPV, F1, FMI, and JI), regardless of CNV size (all, CNV < 100kb, 100kb < CNV < 500kb, or 500kb < CNV < 1Mb), CNV type (all, deletions, or duplications), the highest median performance occurred with Distance Method (Supplemental Figures S38-42). This was particularly evident in the PPV metrics, where Distance Method outperformed BAF, and substantially outperformed LRR mean and LRR sd metrics, except in the 500kb < CNV < 1Mb bin where all four methods appeared to perform about the same (Supplemental Figure S39). The mean-differences between PPV values, unstratified by CNV type or CNV size, of Distance Method , and LRR sd and LRR mean were nominally significant in the Welch two sample t-test ( p = 0.006 and p = 0.027, for LRR sd and LRR mean, respectively). However, no significant differences were observed between any two Method ’s metrics, for any CNV type and CNV size strata after FDR correction (Supplemental Table S1K). We thus opted to examine all Method parameters in the Cross-Array Experiment (CAE). Cross-Array Experiment (CAE) Cross-Array Experiment (CAE) results are summarized in Tables 5 and 6 , Supplemental Tables S1L-N. View this table: View inline View popup Table 5. Summaries of MarkerMatch CNV callsets at D MAX = 10kb across all Method parameters (BAF, LRR mean, LRR sd, and Distance), as well as Full Set and Exact Match reference comparisons for GSA array ( Ref = OEE). Full tables including specific size bins, as well as other statistics like medians and IQRs, are available in the Supplemental Tables S1L-N. Note: callset PPVs are based on comparisons to OEE array Full Set callset. In the QC column, low stands for low-stringency QC and med stands for medium-stringency QC. View this table: View inline View popup Table 6. Summaries of MarkerMatch CNV callsets at D MAX = 10kb across all Method parameters (BAF, LRR mean, LRR sd, and Distance), as well as Full Set and Exact Match reference comparisons for OEE array ( Ref = GSA). Full tables including specific size bins, as well as other statistics like medians and IQRs, are available in the Supplemental Tables S1L-N. Note: callset PPVs are based on comparisons to GSA array Full Set callset. In the QC column, low stands for low-stringency QC and med stands for medium-stringency QC. Briefly, similarly to WAE, Full Sets in both the OEE and GSA array resulted in the most CNV calls (low-stringency QC 24,223 calls and 21,399 calls; medium-stringency QC 4,143 and 1,974 calls on GSA and OEE, respectively) and Exact Match the fewest (low-stringency QC on GSA array 2,529 and 1,883 calls; medium-stringency QC 1,258 and 936 calls on GSA and OEE, respectively). MarkerMatched callsets counted 2-6 times more CNV calls relative to Exact Match across all four Methods at D MAX = 10kb and both arrays (low-stringency QC range 10,491 - 14,024 calls; medium-stringency QC range 1,802 - 2,417 calls). While the average number of low-stringency QC calls overall was not substantially different between the OEE and GSA arrays (GSA overall callsets counted up to 1.3 times more CNVs), the differences between the two arrays observed in medium-stringency QC callsets were more variable and extreme (GSA overall callsets identified up to 2 times more CNVs). CAE per-sample CNV calls, summarized in Tables 5 and 6 , Supplemental Table S1L, were highest for the Full Set callset (low-stringency QC 36.3 and 32.1; medium stringency QC 6.2 and 3.0 for the GSA and OEE arrays, respectively), lowest for Exact Match (low-stringency QC 3.8 and 2.8; medium-stringency QC 1.9 and 1.4 for the GSA and OEE arrays, respectively), and about 2-6 times the Exact Match in MarkerMatched callsets across all Methods at D MAX = 10kb (low-stringency QC ranges of 20.9 - 21.0 and 15.7 - 17.7; medium-stringency QC ranges of 3.6 - 3.6 and 2.7 - 3.5 for GSA and OEE arrays, respectively). In CAE, the average CNV sizes were smaller on the denser configurations ( Tables 5 and 6 , Supplemental Table S1L). The averages CNV sizes for GSA were smallest for MarkerMatch callsets (low-stringency QC range 44.6kb - 44.9kb; medium-stringency QC range 116.1kb - 118.0kb), followed by Full Set (low-stringency QC 58.5kb; medium-stringency QC 138.4kb). The average CNV sizes for OEE were smallest for Full Set (low-stringency QC 40.1kb; medium-stringency QC 112.1kb), followed by MarkerMatch callsets (low-stringency QC range 57.9kb - 62.7kb; medium-stringency QC range 115.0kb - 119.8b). Exact Match in both arrays had the largest average CNV sizes (low-stringency QC 92.1kb and 104.8kb; medium-stringency QC 134.2kb and 138.3kb for GSA and OEE arrays, respectively). Average CNV confidence scores for the GSA array were about the same across all low-stringency QC configurations ( Tables 5 and 6 , Supplemental Table S1L). This included Full Set , Exact Match , and various Method configurations of MarkerMatched callsets (range 22.6 - 22.8), however, in the medium-stringency QC, MarkerMatched callsets had the highest CNV confidence scores (range 54.6 - 55.6), followed by Full Set (51.0) and Exact Match (33.8). For the OEE array, the spread was a bit wider among the low-stringency QC callsets (range 27.9 - 32.2), without clear segregation among the Full Set and MarkerMatch callsets, with Exact Match still being lowest. For medium-stringency QC callsets in the OEE array, we saw the highest average CNV confidence scores for Full Set (99.3), followed by MarkerMatch (range 70.0 - 77.6), and Exact Match (36.8). Overall number of samples with at least one CNV call were consistent after low-stringency QC callsets across the board in both GSA and OEE array (N = 677, Tables 5 and 6 , Supplemental Table S1M). The number of samples passing the medium-stringency QC varied somewhat across the callsets (N ranges 549 - 582 and 400 - 608, for GSA and OEE, respectively), with the OEE array showing a much higher range after medium-stringency QC. Sample-specific metrics did not vary substantially across low-or medium-stringency QC, or Full Set/Exact Match /MarkerMatched configurations, or GSA/OEE arrays ( Tables 5 and 6 , Supplemental Table S1M). Genome-Wide Validation CAE genome-wide validation of the GSA callsets (using low-stringency Full Set OEE as the truth set) resulted in PPVs that were the lowest for the Full Set (0.19), highest in Exact Match (0.56), and intermediate in the MarkerMatched (range 0.295 - 0.298) low-stringency QC callsets ( Figure 5A , Table 5 , Supplemental Table S1N). In medium-stringency QC GSA callsets, the performance was significantly lower in Full Set (0.48) than the rest of the callsets (range 0.72 - 0.74). CAE genome-wide validation of OEE callsets (using low-stringency Full Set GSA as truth set) has resulted in PPVs that were comparable to those in GSA callsets, however overall slightly larger, by an average factor of 1.2 (range 1.1 - 1.6), however Methods LRR mean and LRR sd had a slightly better performance than Exact Match in medium-stringency QC callsets (PPVs of 0.84, 0.84, and 0.82, respectively). Visual inspection of the genome-wide PPV plots ( Figure 5 ) indicates that, in the majority of Method parameter configurations, MarkerMatch performed about as well as the Exact Match approach. Download figure Open in new tab Figure 5. Cross-Array Experiment (CAE) positive predictive value (PPV) plots for both deletions and duplications. Panel A represents calls in Global Screening Array (GSA) validated in Omni Express Exome (OEE). Panel B represents calls in OEE validated in GSA. Empty triangles and dashed lines represent low-stringency QC callsets, whereas solid lines and filled triangles represent medium-stringency QC callsets. Other metrics performed similarly, with larger CNVs having consistently better performance than smaller CNVs (Supplemental Figures S43-48, Supplemental Table S1N). Additionally, duplications had a noticeably better performance than deletions (Supplemental Figure S49-62, Supplemental Table S1N). Regional Validation Filtering on medium-stringency QC indicated that Exact Match outperformed MarkerMatch in variable genomic regions including telomeric, centromeric, and segmental duplication regions (Supplemental Figures S63-69, Supplemental Table S1O) in terms of PPV. However, when accounting for the higher sensitivity of the MarkerMatch using the F1, FMI, and JI metrics, MarkerMatch either matched or slightly outperformed Exact Match across the board. Determination of Optimal Minimum CNV Size and SNP Coverage Thresholds Beta regression of PPV on the GSA array resulted in significant associations with CNV length cutoff, marker coverage cutoff, and their interaction terms ([OR PPV = 1.07, p PPV < 0.001]; [OR PPV = 1.02, p PPV < 0.001]; [OR PPV = 1.00, p PPV < 0.001] respectively). In terms of PPV, LRR sd ([OR PPV = 0.97, p PPV = 0.009]) and BAF ([OR PPV = 0.98, p PPV = 0.03]) seemed to underperform LRR mean. The Distance method was not significantly different. The model’s pseudo R2 was 0.58. In OEE, somewhat counterintuitively, the marker coverage cutoff seemed to actually decrease the PPV ([ OR ppv = 0.98, p PPV < 0.001]) whereas the CNV length cutoff and the two terms’ interaction seemed to increase it ([ OR PPV = 1.02, p PPV < 0.001]; [ OR PPV = 1.00, p PPV < 0.001] respectively). Unlike with sensitivity, Method terms in PPV were significant in OEE array, with LRR sd overperforming ([ OR PPV = 1.07, p PPV = 0.007]), and BAF underperforming ([ OR PPV = 0.91, p PPV < 0.001]) relative to LRR mean. Distance was not significantly different from LRR mean. The model’s pseudo R 2 was 0.64. Graphical representations are shown in Supplemental Figures S70-75. Sample-Wise Performance The plots of performance by sample suggested that a substantial number of false positive calls were driven by poorly-performing samples ( Figure 6 ). For example, considering Exact Match , 113 individuals with PPV < 0.1 accounted for 20.2% of false positive calls in the SSC dataset, 39 individuals with PPV < 0.1 accounted for 22.6% of false positive calls in GSA, and 29 individuals with PPV < 0.1 accounted for 32.5% of false positive calls in OEE (Table S2A, Supplemental Table S1P-Q). Because F1 scores are affected by the changes in sensitivity, and by the QC process in particular, we did not observe similar trends when examining F1 scores (Table S2B). Download figure Open in new tab Figure 6. Figure showing proportions of true positive (blue) to false positive (red) CNV calls across medium-stringency QC callsets for Cross-Array Experiment (Global Screening Array and Omni Express Exome array on columns A and B, respectively) and Within-Array Experiment (Omni2.5 array on panel C). Y-axis shows positive predictive value (PPV), x-axis shows samples (ordered in descending PPV). Plotting the curves along this sample-wise PPV thresholding indicated that, overall, conducting this CNV sample QC step may substantially improve callset PPV (Supplemental Figure S76), however, thresholds that are conservative may result in excessive drop-offs in F1 scores due to associated reductions in sensitivity (Supplemental Figure S77). Discussion We describe a new approach to CNV pre-calling quality control to increase sensitivity in cross-array CNV studies. Instead of exclusively using consensus markers across all arrays considered (an intersection of common markers, or an approach that we call Exact Match in this study), we postulated that using markers in the same genomic neighborhood of the reference marker should result in an identical CNV state call. This ability to rely on similar genomic regions as opposed to identical markers would thus result in improved sensitivity of CNV calling by allowing higher coverage of available array markers ( Figure 3 ). Using simulated (within array experiment; WAE) data from the OMNI array, we show that using MarkerMatch not only substantially increased the sensitivity of CNV calling (4-fold increase in sensitivity from 0.031 in Exact Match to average sensitivity of 0.124 in MarkerMatch approach), it did so without a negative impact to PPV (0.900 in Exact Match compared to 0.943-958 in MarkerMatch) shown in Figure 4 and Table 4 . Fluctuations in apparent performance are dependent on the number of CNVs in the truth set, for example, in specific subsets of CNV callsets such as CNV size cutoff > 5Mb only have between 2 and 25 CNV calls, therefore, a single false positive call may greatly affect observed PPV ( Figure 4F ). We identified a 10,000bp D MAX< parameter as a reasonable setting, and note that none of the specific Method parameters had a clear performance edge over the others. Noticeably, however, performance metrics that take into account both sensitivity and PPV, such as F1 score, Fowlkes-Mallows index, and Jaccard index, showed MarkerMatch substantially overperforming Exact Match (Supplemental Figures S9-11). In the cross array experiment (CAE), we used TAAICG samples for which data were generated on two different arrays, and show that both Exact Match and MarkerMatch reduced some of the batch effects associated with the use of different arrays (PPV range of Full Set callset 0.479 vs. 0.705 - 0.741 in Exact Match and MarkerMatch callsets). CAE also demonstrated that MarkerMatch performed about as well as or better than Exact Match in terms of PPV ( Figure 5 , Tables 5 and 6 ). Similar to the WAE, performance metrics that take into account both sensitivity and PPV, such as F1, FMI, and JI, showed MarkerMatch outperforming Exact Match (Supplemental Figures S46-48). We also determined that increases in CNV length and marker coverage cutoffs drive improvements in PPV, but may cause reduction in overall sensitivity, as expected (Supplemental Figures S70-75). We further inspected sample-wise performance rates to determine whether low PPVs were driven by individuals with low PPV ( Figure 6 ), with individual samples with sample-wise PPV < 0.1 (ranging 0-14.4% of samples in a given callset) accounting for 20.2-32.5% of false positive CNV calls. We further examined whether eliminating these samples would substantially improve quality of the callsets, and found that while removing them may lead to noticeable improvements in PPV by a few percentage points, excessively conservative (high) thresholding hurt sensitivity and overall performance as measured by F1 scores (Supplemental Figures S76 and S77). Notably, excluding samples with low PPV values (< 0.1) does not decrease sensitivity or F1 scores. While the MarkerMatch approach is successful in its primary function by increasing/rescuing sensitivity without reducing/sacrificing PPV, MarkerMatch does not eliminate batch effects attributable to the use of different arrays. We found some evidence of the potential reduction in batch and array effects in this study, but this needs to be further explored. Further research needs to be done to determine how significant array-specific batch effects really are, how much do Exact Match or MarkerMatch approaches really alleviate them, and what their quantifiable consequence to downstream CNV analyses might be. Thus, it is noteworthy that batch and array effects, albeit demonstrably reduced by MarkerMatch, still remain an important consideration in downstream CNV association analyses. Additionally, because we lacked access to adequate ancestrally diverse data, we did not examine the effects of ancestry composition on the MarkerMatch algorithm. Funding This work was supported by the National Institute for Neurological Disorders and Stroke [NS082168 to F.I.; NS105746 to R.O, L.D., J.S., and C.M.; NS102371 to R.O., L.D., J.S., and C.M.; MH115676 and MH125042 to R.O.; Tourette Association of America Young Investigator Award to T.M.F.]. Acknowledgements The authors would like to thank the Simons Foundation for Autism Research and the Tourette Association of America International Consortium for Genetics for contributing data for the analysis. The authors would also like to thank professor Lea Davis and Sharon Johnson for their feedback in the project. Funder Information Declared National Institute of Neurological Disorders and Stroke, https://ror.org/01s5ya894 , NS102371 , NS105746 , NS082168 National Institute of Mental Health, https://ror.org/04xeg9z08 , MH125042 , MH115676 Tourette Association of America, https://ror.org/00x2ve368 Footnotes Availability and Implementation The complete code and implementation for the MarkerMatch algorithm and all analyses in this paper are available at: https://github.com/FranjoIM/MarkerMatch . Supplementary Information Supplementary data are available online. Data Availability Statement The data underlying this article were provided by the Simons Foundation for Autism Research and the Tourette Association of America International Consortium for Genetics under license / by permission. Data can be accessed through the Simons Foundation for Autism Research and the Tourette Association of America International Consortium for Genetics. https://github.com/FranjoIM/MarkerMatch . Abbreviations ASD Autism spectrum disorder BAF B allele frequency CNV Copy-number variant D_MAX Maximum (allowable) distance F1 F1 score FDR False discovery rate FMI Fowlkes–Mallows index GSA Global Screening Array LRR Log R ratio MAD Median absolute deviation OEE Omni Express Exome OMNI Omni 2.5 PPV Positive predictive value QC Quality control SSC Simons Simplex Cohort TAAICG Tourette Association of America International Consortium for Genetics TS Tourette syndrome References Bengtsson H , Corrada Bravo H , Gentleman R et al. matrixStats: Functions that Apply to Rows and Columns of Matrices (and to Vectors) . 2017 . ↵ Bycroft C , Freeman C , Petkova D et al. The UK Biobank resource with deep phenotyping and genomic data . Nature 2018 ; 562 : 203 – 9 . OpenUrl CrossRef PubMed ↵ Colella S , Yau C , Taylor JM et al. QuantiSNP: an Objective Bayes Hidden-Markov Model to detect and accurately map copy number variation using SNP genotyping data . Nucleic Acids Res 2007 ; 35 : 2013 – 25 . OpenUrl CrossRef PubMed Web of Science Cribari-Neto F , Zeileis A . Beta Regression in R . J Stat Softw 2010 ; 34 , DOI: 10.18637/jss.v034.i02 . OpenUrl CrossRef Diskin SJ , Li M , Hou C et al. Adjustment of genomic waves in signal intensities from whole-genome SNP genotyping platforms . Nucleic Acids Res 2008 ; 36 : e126 – e126 . OpenUrl CrossRef PubMed ↵ Ehli EA , Abdellaoui A , Fedko IO et al. A method to customize population-specific arrays for genome-wide association testing . Eur J Hum Genet 2017 ; 25 : 267 – 70 . OpenUrl CrossRef PubMed ↵ Fischbach GD , Lord C . The Simons Simplex Collection: A Resource for Identification of Autism Genetic Risk Factors . Neuron 2010 ; 68 : 192 – 5 . OpenUrl CrossRef PubMed Web of Science ↵ Fu JM , Satterstrom FK , Peng M et al. Rare coding variation provides insight into the genetic architecture and phenotypic context of autism . Nat Genet 2022 ; 54 : 1320 – 31 . OpenUrl CrossRef PubMed Harris CR , Millman KJ , Van Der Walt SJ et al. Array programming with NumPy . Nature 2020 ; 585 : 357 – 62 . OpenUrl CrossRef PubMed ↵ Huang AY , Yu D , Davis LK et al. Rare Copy Number Variants in NRXN1 and CNTN6 Increase Risk for Tourette Syndrome . Neuron 2017 ; 94 : 1101 – 1111 .e7. OpenUrl CrossRef PubMed Illumina. DNA Copy Number and Loss of Heterozygosity Analysis Algorithms . 2017 . Illumina. GenomeStudio . 2020 . Kassambara A. ggpubr: “ggplot2” Based Publication Ready Plots. 2023 . ↵ Lavrichenko K , Helgeland Ø , Njølstad PR et al. SeeCiTe: a method to assess CNV calls from SNP arrays using trio data. Schwartz R (ed.) . Bioinformatics 2021 ; 37 : 1876 – 83 . OpenUrl CrossRef PubMed ↵ Lionel AC , Crosbie J , Barbosa N et al. Rare Copy Number Variation Discovery and Cross-Disorder Comparisons Identify Risk Genes for ADHD . Sci Transl Med 2011 ; 3 , DOI: 10.1126/scitranslmed.3002464 . OpenUrl Abstract / FREE Full Text ↵ Marchini J , Howie B . Genotype imputation for genome-wide association studies . Nat Rev Genet 2010 ; 11 : 499 – 511 . OpenUrl CrossRef PubMed Web of Science ↵ Marshall CR , Psychosis Endophenotypes International Consortium, CNV and Schizophrenia Working Groups of the Psychiatric Genomics Consortium et al. Contribution of copy number variants to schizophrenia from a genome-wide study of 41,321 subjects . Nat Genet 2017 ; 49 : 27 – 35 . OpenUrl CrossRef PubMed ↵ McCarroll SA , Kuruvilla FG , Korn JM et al. Integrated detection and population-genetic analysis of SNPs and copy number variation . Nat Genet 2008 ; 40 : 1166 – 74 . OpenUrl CrossRef PubMed Web of Science ↵ Nakatochi M , Kushima I , Ozaki N . Implications of germline copy-number variations in psychiatric disorders: review of large-scale genetic studies . J Hum Genet 2021 ; 66 : 25 – 37 . OpenUrl CrossRef PubMed ↵ Olson H , Shen Y , Avallone J et al. Copy number variation plays an important role in clinical epilepsy . Ann Neurol 2014 ; 75 : 943 – 58 . OpenUrl CrossRef PubMed ↵ Pinto D , Darvishi K , Shi X et al. Comprehensive assessment of array-based platforms and calling algorithms for detection of copy number variants . Nat Biotechnol 2011 ; 29 : 512 – 20 . OpenUrl CrossRef PubMed Web of Science Posit Team. RStudio: Integrated Development Environment for R. 2024 . Purcell S , Neale B , Todd-Brown K et al. PLINK: A Tool Set for Whole-Genome Association and Population-Based Linkage Analyses . Am J Hum Genet 2007 ; 81 : 559 – 75 . OpenUrl CrossRef PubMed Python Software Foundation. Python Language Reference. 2023 . R Core Team. R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing. 2021 . ↵ Sanders SJ , He X , Willsey AJ et al. Insights into Autism Spectrum Disorder Genomic Architecture and Biology from 71 Risk Loci . Neuron 2015 ; 87 : 1215 – 33 . OpenUrl CrossRef PubMed ↵ Scharf JM , Yu D , Mathews CA et al. Genome-wide association study of Tourette’s syndrome . Mol Psychiatry 2013 ; 18 : 721 – 8 . OpenUrl CrossRef PubMed ↵ Scheet P , Stephens M . A Fast and Flexible Statistical Model for Large-Scale Population Genotype Data: Applications to Inferring Missing Genotypes and Haplotypic Phase . Am J Hum Genet 2006 ; 78 : 629 – 44 . OpenUrl CrossRef PubMed Web of Science ↵ Sun W , Wright FA , Tang Z et al. Integrated study of copy number states and genotype calls using high-density SNP arrays . Nucleic Acids Res 2009 ; 37 : 5365 – 77 . OpenUrl CrossRef PubMed Web of Science The All of Us Research Program Genomics Investigators, Manuscript Writing Group, Bick AG et al. Genomic data in the All of Us Research Program . Nature 2024 ; 627 : 340 – 6 . OpenUrl CrossRef PubMed The pandas development team. pandas-dev/pandas: Pandas. 2024 , DOI: 10.5281/ZENODO.13819579 . OpenUrl CrossRef Wang K , Chen Z , Tadesse MG et al. Modeling genetic inheritance of copy number variations . Nucleic Acids Res 2008 ; 36 : e138 – e138 . OpenUrl CrossRef PubMed ↵ Wang K , Li M , Hadley D et al. PennCNV: An integrated hidden Markov model designed for high-resolution copy number variation detection in whole-genome SNP genotyping data . Genome Res 2007 ; 17 : 1665 – 74 . OpenUrl Abstract / FREE Full Text ↵ Wang S , Mandell JD , Kumar Y et al. De Novo Sequence and Copy Number Variants Are Strongly Associated with Tourette Disorder and Implicate Cell Polarity in Pathogenesis . Cell Rep 2018 ; 24 : 3441 – 3454 .e12. OpenUrl CrossRef PubMed ↵ Wickham H , Averick M , Bryan J et al. Welcome to the Tidyverse . J Open Source Softw 2019 ; 4 : 1686 . OpenUrl CrossRef ↵ Zarrei M , MacDonald JR , Merico D et al. A copy number variation map of the human genome . Nat Rev Genet 2015 ; 16 : 172 – 83 . OpenUrl CrossRef PubMed ↵ Zeggini E , Ioannidis JP . Meta-Analysis in Genome-Wide Association Studies . Pharmacogenomics 2009 ; 10 : 191 – 201 . OpenUrl CrossRef PubMed Web of Science ↵ Zhang Z , Cheng H , Hong X et al. EnsembleCNV: an ensemble machine learning algorithm to identify and genotype copy number variation using SNP array data . Nucleic Acids Res 2019 ; 47 : e39 – e39 . OpenUrl CrossRef PubMed View the discussion thread. Back to top Previous Next Posted July 04, 2025. Download PDF Supplementary Material Data/Code Email Thank you for your interest in spreading the word about bioRxiv. NOTE: Your email address is requested solely to identify you as the sender of this article. Your Email * Your Name * Send To * Enter multiple addresses on separate lines or separate them with commas. You are going to email the following MarkerMatch: A Proximity-Based Probe-Matching Algorithm for Joint Analysis of Copy-Number Variants from Different Genotyping Arrays Message Subject (Your Name) has forwarded a page to you from bioRxiv Message Body (Your Name) thought you would like to see this page from the bioRxiv website. Your Personal Message CAPTCHA This question is for testing whether or not you are a human visitor and to prevent automated spam submissions. Share MarkerMatch: A Proximity-Based Probe-Matching Algorithm for Joint Analysis of Copy-Number Variants from Different Genotyping Arrays Franjo Ivankovic , Dongmei Yu , James Shen , Lingyu Zhan , Maria Niarchou , Ariadne Kaylor , Laura Domènech , Tyne W Miller-Fleming , Luz M Porras , Paola Giusti-Rodríguez , Roel A Ophoff , Jeremiah M Scharf , Carol A Mathews bioRxiv 2025.06.30.662249; doi: https://doi.org/10.1101/2025.06.30.662249 Share This Article: Copy Citation Tools MarkerMatch: A Proximity-Based Probe-Matching Algorithm for Joint Analysis of Copy-Number Variants from Different Genotyping Arrays Franjo Ivankovic , Dongmei Yu , James Shen , Lingyu Zhan , Maria Niarchou , Ariadne Kaylor , Laura Domènech , Tyne W Miller-Fleming , Luz M Porras , Paola Giusti-Rodríguez , Roel A Ophoff , Jeremiah M Scharf , Carol A Mathews bioRxiv 2025.06.30.662249; doi: https://doi.org/10.1101/2025.06.30.662249 Citation Manager Formats BibTeX Bookends EasyBib EndNote (tagged) EndNote 8 (xml) Medlars Mendeley Papers RefWorks Tagged Ref Manager RIS Zotero Tweet Widget Facebook Like Google Plus One Subject Area Bioinformatics Subject Areas All Articles Animal Behavior and Cognition (7618) Biochemistry (17635) Bioengineering (13859) Bioinformatics (41846) Biophysics (21401) Cancer Biology (18534) Cell Biology (25422) Clinical Trials (138) Developmental Biology (13352) Ecology (19860) Epidemiology (2067) Evolutionary Biology (24285) Genetics (15582) Genomics (22463) Immunology (17700) Microbiology (40298) Molecular Biology (17141) Neuroscience (88424) Paleontology (666) Pathology (2825) Pharmacology and Toxicology (4813) Physiology (7633) Plant Biology (15107) Scientific Communication and Education (2042) Synthetic Biology (4284) Systems Biology (9808) Zoology (2267)