Portal de Periódicos da CAPES

Somatic Tumor Variant Filtration Strategies to Optimize Tumor-Only Molecular Profiling Using Targeted Next-Generation Sequencing Panels

2018; Elsevier BV; Volume: 21; Issue: 2 Linguagem: Inglês

10.1016/j.jmoldx.2018.09.008

1943-7811

Mahadeo A. Sukhai, Maksym Misyura, Mariam Thomas, Swati Garg, Tong Zhang, Natalie Stickle, Carl Virtanen, Philippe L. Bédard, Lillian L. Siu, Tina Smets, Gert Thijs, Steven Van Vooren, Suzanne Kamel‐Reid, Tracy Stockley,

CRISPR and Genetic Engineering

A common approach in clinical diagnostic laboratories to variant assessment from tumor molecular profiling is sequencing of genomic DNA extracted from both tumor (somatic) and normal (germline) tissue, with subsequent variant comparison to identify true somatic variants with potential impact on patient treatment or prognosis. However, challenges exist in paired tumor-normal testing, including increased cost of dual sample testing and identification of germline cancer predisposing variants. Alternatively, somatic variants can be identified by in silico tumor-only variant filtration precluding the need for matched normal testing. The barrier to tumor-only variant filtration is defining a reliable approach, with high sensitivity and specificity to identify somatic variants. In this study, we used retrospective data sets from paired tumor-normal samples tested on small (48 gene) and large (555 gene) targeted next-generation sequencing panels, to model algorithms for tumor-only variants classification. The optimal algorithm required an ordinal filtering approach using information from variant population databases (1000 Genomes Phase 3, ESP6500, ExAC), clinical mutation databases (ClinVar), and information on recurring clinically relevant somatic variants. Overall the tumor-only variant filtration strategy described in this study can define clinically relevant somatic variants from tumor-only analysis with sensitivity of 97% to 99% and specificity of 87% to 94%, and with significant potential utility for clinical laboratories implementing tumor-only molecular profiling. A common approach in clinical diagnostic laboratories to variant assessment from tumor molecular profiling is sequencing of genomic DNA extracted from both tumor (somatic) and normal (germline) tissue, with subsequent variant comparison to identify true somatic variants with potential impact on patient treatment or prognosis. However, challenges exist in paired tumor-normal testing, including increased cost of dual sample testing and identification of germline cancer predisposing variants. Alternatively, somatic variants can be identified by in silico tumor-only variant filtration precluding the need for matched normal testing. The barrier to tumor-only variant filtration is defining a reliable approach, with high sensitivity and specificity to identify somatic variants. In this study, we used retrospective data sets from paired tumor-normal samples tested on small (48 gene) and large (555 gene) targeted next-generation sequencing panels, to model algorithms for tumor-only variants classification. The optimal algorithm required an ordinal filtering approach using information from variant population databases (1000 Genomes Phase 3, ESP6500, ExAC), clinical mutation databases (ClinVar), and information on recurring clinically relevant somatic variants. Overall the tumor-only variant filtration strategy described in this study can define clinically relevant somatic variants from tumor-only analysis with sensitivity of 97% to 99% and specificity of 87% to 94%, and with significant potential utility for clinical laboratories implementing tumor-only molecular profiling. A significant challenge in tumor-only molecular profiling is defining somatic tumor-specific variants in the background of germline variants also detected during sequencing. To distinguish somatic tumor variants from germline variants, two major approaches have emerged: parallel testing of a normal germline DNA sample, typically from blood or adjacent normal tissue from formalin-fixed, paraffin-embedded (FFPE) samples, with subtraction of germline variants from the total variants identified in the tumor tissue to produce a list of somatic tumor-only variants1Schrader K.A. Cheng D.T. Joseph V. Prasad M. Walsh M. Zehir A. Ni A. Thomas T. Benayed R. Ashraf A. Lincoln A. Arcila M. Stadler Z. Solit D. Hyman D.M. Zhang L. Klimstra D. Ladanyi M. Offit K. Berger M. Robson M. Germline variants in targeted tumor sequencing using matched normal DNA.JAMA Oncol. 2016; 2: 104-111Crossref PubMed Scopus (207) Google Scholar, 2Jones S. Anagnostou V. Lytle K. Parpart-Li S. Nesselbush M. Riley D.R. Shukla M. Chesnick B. Kadan M. Papp E. Galens K.G. Murphy D. Zhang T. Kann L. Sausen M. Angiuoli S.V. Diaz Jr., L.A. Velculescu V.E. Personalized genomic analyses for cancer mutation discovery and interpretation.Sci Transl Med. 2015; 7: 283ra53Crossref PubMed Scopus (294) Google Scholar or tumor-only analysis, with in silico variant filtration using available databases and other resources to prioritize variants likely to be somatic.3Hiltemann S. Jenster G. Trapman J. van der Spek P. Stubbs A. Discriminating somatic and germline mutations in tumor DNA samples without matching normals.Genome Res. 2015; 25: 1382-1390Crossref PubMed Scopus (56) Google Scholar In the clinical molecular diagnostic laboratory, parallel testing of germline samples to classify somatic tumor variants by tumor-normal comparison raises issues for practice. The additional expense of testing both the germline and tumor sample from each patient is significant. In addition, the potential identification of inherited germline cancer predisposing variants in normal samples requires appropriate management, including pretest consent, germline variant–specific investigations, and appropriate return of results with genetic counseling support.4Robson M.E. Bradbury A.R. Arun B. Domchek S.M. Ford J.M. Hampel H.L. Lipkin S.M. Syngal S. Wollins D.S. Lindor N.M. American Society of Clinical Oncology Policy Statement Update: genetic and genomic testing for cancer susceptibility.J Clin Oncol. 2015; 33: 3660-3667Crossref PubMed Scopus (390) Google Scholar Tumor-only testing for identification of somatic variants involves in silico filtering of variants against multiple information sources, including laboratory or online databases that contain information on somatic or germline variants, published literature, and protein prediction tools. Collectively, the information in these databases is limited by variable accuracy, hindering the approach.5Yen J.L. Garcia S. Montana A. Harris J. Chervitz S. Morra M. West J. Chen R. Church D.M. A variant by any name: quantifying annotation discordance across tools and clinical databases.Genome Med. 2017; 9: 7Crossref PubMed Scopus (38) Google Scholar, 6Harrison S.M. Dolinsky J.S. Knight Johnson A.E. Pesaran T. Azzariti D.R. Bale S. Chao E.C. Das S. Vincent L. Rehm H.L. Clinical laboratories collaborate to resolve differences in variant interpretations submitted to ClinVar.Genet Med. 2017; 19: 1096-1104Abstract Full Text Full Text PDF PubMed Scopus (149) Google Scholar, 7Harrison S.M. Riggs E.R. Maglott D.R. Lee J.M. Azzariti D.R. Niehaus A. Ramos E.M. Martin C.L. Landrum M.J. Rehm H.L. Using ClinVar as a resource to support variant interpretation.Curr Protoc Hum Genet. 2016; 89: 8.16.1-8.16.23Crossref Scopus (77) Google Scholar For example, germline polymorphic variant databases, such as Single Nucleotide Polymorphism Database (dbSNP), contain pathogenic germline variants,8Nishiguchi K.M. Tearle R.G. Liu Y.P. Oh E.C. Miyake N. Benaglio P. Harper S. Koskiniemi-Kuendig H. Venturini G. Sharon D. Koenekoop R.K. Nakamura M. Kondo M. Ueno S. Yasuma T.R. Beckmann J.S. Ikegawa S. Matsumoto N. Terasaki H. Berson E.L. Katsanis N. Rivolta C. Whole genome sequencing in patients with retinitis pigmentosa reveals pathogenic DNA structural changes and NEK2 as a new disease gene.Proc Natl Acad Sci U S A. 2013; 110: 16139-16144Crossref PubMed Scopus (99) Google Scholar, 9Arthur J.W. Cheung F.S.G. Reichardt J.K.V. Single nucleotide differences (SNDs) continue to contaminate the dbSNP database with consequences for human genomics and health.Hum Mutat. 2015; 36: 196-199Crossref PubMed Scopus (8) Google Scholar, 10Sherry S.T. Ward M.H. Kholodov M. Baker J. Phan L. Smigielski E.M. Sirotkin K. dbSNP: the NCBI database of genetic variation.Nucleic Acids Res. 2001; 29: 308-311Crossref PubMed Scopus (4865) Google Scholar whereas somatic variant databases, such as Catalogue of Somatic Mutations in Cancer (COSMIC), contain germline variants.5Yen J.L. Garcia S. Montana A. Harris J. Chervitz S. Morra M. West J. Chen R. Church D.M. A variant by any name: quantifying annotation discordance across tools and clinical databases.Genome Med. 2017; 9: 7Crossref PubMed Scopus (38) Google Scholar A recent publication assessed whole exome sequencing data filtering to classify variants with the use of paired and unpaired approaches. The study defined an improved set of criteria for successful discrimination of somatic and germline variants; however, optimal sensitivity and specificity were achieved by use of a paired normal sample.11Garofalo A. Sholl L. Reardon B. Taylor-Weiner A. Amin-Mansour A. Miao D. Liu D. Oliver N. MacConaill L. Ducar M. Rojas-Rudilla V. Giannakis M. Ghazani A. Gray S. Janne P. Garber J. Joffe S. Lindeman N. Wagle N. Garraway L.A. Van Allen E.M. The impact of tumor profiling approaches and genomic data strategies for cancer precision medicine.Genome Med. 2016; 8: 79Crossref PubMed Scopus (124) Google Scholar To minimize the potentially significant impact on patient care of inaccurately defined somatic variants1Schrader K.A. Cheng D.T. Joseph V. Prasad M. Walsh M. Zehir A. Ni A. Thomas T. Benayed R. Ashraf A. Lincoln A. Arcila M. Stadler Z. Solit D. Hyman D.M. Zhang L. Klimstra D. Ladanyi M. Offit K. Berger M. Robson M. Germline variants in targeted tumor sequencing using matched normal DNA.JAMA Oncol. 2016; 2: 104-111Crossref PubMed Scopus (207) Google Scholar, 2Jones S. Anagnostou V. Lytle K. Parpart-Li S. Nesselbush M. Riley D.R. Shukla M. Chesnick B. Kadan M. Papp E. Galens K.G. Murphy D. Zhang T. Kann L. Sausen M. Angiuoli S.V. Diaz Jr., L.A. Velculescu V.E. Personalized genomic analyses for cancer mutation discovery and interpretation.Sci Transl Med. 2015; 7: 283ra53Crossref PubMed Scopus (294) Google Scholar and to improve in silico assessment of somatic variants, we present development and validation of an optimized tumor-only variant filtration strategy for targeted panel tumor molecular profiling. The variant filtration algorithm was developed by using variant results from a medium-sized (48 gene) targeted next-generation sequencing (NGS) panel tested on 1120 tumor FFPE and matched blood samples. The tumor-only filtering algorithm was also tested on a variant data set from 53 pairs of samples (tumor FFPE and blood samples from the same patients) tested on a large (555 genes) targeted NGS panel. The tumor-only filtering algorithm was found to have 99% sensitivity and 94% specificity for detection of somatic variants from medium NGS panels, and 97% sensitivity and 87% specificity from large NGS panels, without the need for testing of matched normal samples. The impact of including commonly used germline and somatic databases and other tools was also modeled on variant filtration, and the optimal variant filtration approach defined for targeted NGS panel tumor-only variant analysis. Variant data were generated from DNA samples extracted from FFPE tumor specimens (biopsies or surgical resections) or peripheral blood lymphocytes (PBLs). Samples were collected under a University Health Network Research Ethics Board–approved study as previously described.12Stockley T.L. Oza A.M. Berman H.K. Leighl N.B. Knox J.J. Shepherd F.A. Chen E.X. Krzyzanowska M.K. Dhani N. Joshua A.M. Tsao M.-S. Serra S. Clarke B. Roehrl M.H. Zhang T. Sukhai M.A. Califaretti N. Trinkaus M. Shaw P. van der Kwast T. Wang L. Virtanen C. Kim R.H. Razak A.R.A. Hansen A.R. Yu C. Pugh T.J. Kamel-Reid S. Siu L.L. Bedard P.L. Molecular profiling of advanced solid tumors and patient outcomes with genotype-matched clinical trials: the Princess Margaret IMPACT/COMPACT trial.Genome Med. 2016; 8: 109Crossref PubMed Scopus (156) Google Scholar Tumor regions of FFPE specimens were acceptable if tumor cellularity was ≥20%, and tumor isolated by 1 to 2 × 1 mm punch from FFPE blocks or macrodissection of unstained material from 15 to 20 slides (4 to 7 μm sections), and DNA extracted from FFPE or PBLs as described.12Stockley T.L. Oza A.M. Berman H.K. Leighl N.B. Knox J.J. Shepherd F.A. Chen E.X. Krzyzanowska M.K. Dhani N. Joshua A.M. Tsao M.-S. Serra S. Clarke B. Roehrl M.H. Zhang T. Sukhai M.A. Califaretti N. Trinkaus M. Shaw P. van der Kwast T. Wang L. Virtanen C. Kim R.H. Razak A.R.A. Hansen A.R. Yu C. Pugh T.J. Kamel-Reid S. Siu L.L. Bedard P.L. Molecular profiling of advanced solid tumors and patient outcomes with genotype-matched clinical trials: the Princess Margaret IMPACT/COMPACT trial.Genome Med. 2016; 8: 109Crossref PubMed Scopus (156) Google Scholar DNA samples (n = 1120) from FFPE tumor tissue (Supplemental Figure S1) (described in Stockley et al12Stockley T.L. Oza A.M. Berman H.K. Leighl N.B. Knox J.J. Shepherd F.A. Chen E.X. Krzyzanowska M.K. Dhani N. Joshua A.M. Tsao M.-S. Serra S. Clarke B. Roehrl M.H. Zhang T. Sukhai M.A. Califaretti N. Trinkaus M. Shaw P. van der Kwast T. Wang L. Virtanen C. Kim R.H. Razak A.R.A. Hansen A.R. Yu C. Pugh T.J. Kamel-Reid S. Siu L.L. Bedard P.L. Molecular profiling of advanced solid tumors and patient outcomes with genotype-matched clinical trials: the Princess Margaret IMPACT/COMPACT trial.Genome Med. 2016; 8: 109Crossref PubMed Scopus (156) Google Scholar) and matched DNA samples from PBLs (ie, 2240 samples from 1120 participants) were tested (250 ng of DNA for library preparation) with the targeted hotspot TruSeq Amplicon Cancer Panel (TSCAP; Illumina, San Diego, CA) that covered regions of 48 genes12Stockley T.L. Oza A.M. Berman H.K. Leighl N.B. Knox J.J. Shepherd F.A. Chen E.X. Krzyzanowska M.K. Dhani N. Joshua A.M. Tsao M.-S. Serra S. Clarke B. Roehrl M.H. Zhang T. Sukhai M.A. Califaretti N. Trinkaus M. Shaw P. van der Kwast T. Wang L. Virtanen C. Kim R.H. Razak A.R.A. Hansen A.R. Yu C. Pugh T.J. Kamel-Reid S. Siu L.L. Bedard P.L. Molecular profiling of advanced solid tumors and patient outcomes with genotype-matched clinical trials: the Princess Margaret IMPACT/COMPACT trial.Genome Med. 2016; 8: 109Crossref PubMed Scopus (156) Google Scholar on the MiSeq sequencer (Illumina). The TSACP was selected because it is a well-used and well-validated commercially available targeted panel that was found to yield high-quality sequencing data.12Stockley T.L. Oza A.M. Berman H.K. Leighl N.B. Knox J.J. Shepherd F.A. Chen E.X. Krzyzanowska M.K. Dhani N. Joshua A.M. Tsao M.-S. Serra S. Clarke B. Roehrl M.H. Zhang T. Sukhai M.A. Califaretti N. Trinkaus M. Shaw P. van der Kwast T. Wang L. Virtanen C. Kim R.H. Razak A.R.A. Hansen A.R. Yu C. Pugh T.J. Kamel-Reid S. Siu L.L. Bedard P.L. Molecular profiling of advanced solid tumors and patient outcomes with genotype-matched clinical trials: the Princess Margaret IMPACT/COMPACT trial.Genome Med. 2016; 8: 109Crossref PubMed Scopus (156) Google Scholar The UHN Hi5 was also similarly designed and validated as a high-quality clinical panel. One hundred six matched FFPE-PBL DNA samples (from 53 patients) were also tested by using a validated custom capture 555-gene panel (UHN Hi5) that covered exons and minimum of 10 bp of flanking intronic region of 555 cancer-related genes (Supplemental Table S1). Libraries were constructed from 250 ng of DNA sheared by sonication (Covaris, Woburn, MA), with end repair and ligation with barcoded sequencing adaptors, followed by hybrid capture with RNA baits (SureSelect; Agilent, Santa Clara, CA) and sequencing on the NextSeq (Illumina). For TSCAP, sequence alignment and variant calling for all samples were performed with MiSeq Reporter software version 2.3.1 (Illumina), and variants were reviewed with the Integrative Genomics Viewer (Broad Institute, Cambridge, MA). Somatic variants included for analysis passed MiSeq Reporter quality filter and met laboratory-defined thresholds of ≥250× read depth and >5% variant allele fraction (VAF) in DNA from FFPE tissue. Germline variants included for analysis met thresholds of >50× read depth in DNA from PBLs. Three genes with read depth consistently 2500 solid tumor samples, using the Illumina TSACP, within the UHN AMDL from 2012 to 2016. Classification of variants as clinically relevant was performed by using our previously published approach and classification scheme.16Sukhai M.A. Craddock K.J. Thomas M. Hansen A.R. Zhang T. Siu L. Bedard P. Stockley T.L. Kamel-Reid S. A classification system for clinical relevance of somatic variants identified in molecular profiling of cancer.Genet Med. 2016; 18: 128-136Abstract Full Text Full Text PDF PubMed Scopus (70) Google Scholar The CR-MVL was used in phase I to retain any clinically actionable variant with <5% VAF. In phase II, variants retained after phase I were labeled as germline if they were present in any of the following four germline population variant databases (PVDs) at a minor allele frequency (MAF) of ≥1%: 1000 Genomes phase 3 (release version 5.2013050217Siva N. 1000 Genomes project.Nat Biotechnol. 2008; 26: 256Crossref PubMed Scopus (267) Google Scholar, 18Auton A. Brooks L.D. Durbin R.M. Garrison E.P. Kang H.M. Korbel J.O. Marchini J.L. McCarthy S. McVean G.A. Abecasis G.R. 1000 Genomes Project ConsortiumA global reference for human genetic variation.Nature. 2015; 526: 68-74Crossref PubMed Scopus (8527) Google Scholar); Exome Sequencing Project (ESP; ESP6500SI-V2 data set of Exome Variant Server, National Heart, Lung, and Blood Institute Grand Opportunity Exome Sequencing Project, Seattle, WA; http://evs.gs.washington.edu/EVS, last accessed August 2016), Exome Aggregation Consortium version 0.3 (ExAC)19Lek M. Karczewski K.J. Minikel E.V. Samocha K.E. Banks E. Fennell T. et al.Analysis of protein-coding genetic variation in 60,706 humans.Nature. 2016; 536: 285-291Crossref PubMed Scopus (6602) Google Scholar or dbSNP build 141 (GRCh37.p13).10Sherry S.T. Ward M.H. Kholodov M. Baker J. Phan L. Smigielski E.M. Sirotkin K. dbSNP: the NCBI database of genetic variation.Nucleic Acids Res. 2001; 29: 308-311Crossref PubMed Scopus (4865) Google Scholar Variants not found in any of the PVDs at the ≥1% MAF threshold were labeled as somatic. Also in phase II, variants were labeled as somatic if they were found in the COSMIC database release version 71,20Forbes S.A. Beare D. Gunasekaran P. Leung K. Bindal N. Boutselakis H. Ding M. Bamford S. Cole C. Ward S. Kok C.Y. Jia M. De T. Teague J.W. Stratton M.R. McDermott U. Campbell P.J. COSMIC: exploring the world's knowledge of somatic mutations in human cancer.Nucleic Acids Res. 2015; 43: D805-D811Crossref PubMed Scopus (1766) Google Scholar with two or more occurrences, even if also found in a germline population database. Finally, in phase III, variants were labeled germline if they were present as a benign or likely benign variant in either the Human Gene Mutation Database (HGMD; HGMD Professional Database 2015.1; http://www.hgmd.cf.ac.uk/ac/index.php)21Krawczak M. Ball E.V. Stenson P. Cooper D.N. HGMD: the human gene mutation database.in: Letovsky S. Bioinformatics: Databases and Systems. Springer, Boston, MA2002: 99-104Crossref Google Scholar or in ClinVar (National Center for Biotechnology Information ClinVar; 20150504; https://www.ncbi.nlm.nih.gov/clinvar).22Landrum M.J. Lee J.M. Benson M. Brown G. Chao C. Chitipiralla S. Gu B. Hart J. Hoffman D. Hoover J. Jang W. Katz K. Ovetsky M. Riley G. Sethi A. Tully R. Villamarin-Salomon R. Rubinstein W. Maglott D.R. ClinVar: public archive of interpretations of clinically relevant variants.Nucleic Acids Res. 2016; 44: D862-D868Crossref PubMed Scopus (1538) Google Scholar All variant classifications (germline or somatic) from the tumor-only filtration algorithm were compared with matched tumor-normal sample analysis output, whereby variants occurring in DNA from both the FFPE tumor and the matched normal PBLs were considered germline and removed from the tumor variant list. Tumor-normal analysis was conducted within Alissa Interpret by using a second custom filtration algorithm tree that compared variants from the TSACP variant call files from the PBL and FFPE samples for each case (tumor-normal matched sample algorithm not shown). Comparison of the results from both analyses enabled classification of variants from the tumor-only analysis as true somatic call (TSC), true germline call (TGC), false somatic call (FSC), and false germline call (FGC) in the context of somatic variant classification by the tumor-only variant filtration algorithm (Table 1). Sensitivity, specificity, positive predictive value, and negative predictive value (NPV) (Table 1) were used for performance evaluation of the tumor-only variant filtration algorithm for classification of somatic variants.Table 1Definitions as Applied to Variants Classified by the Tumor-Only Filtration Algorithm and Approach to Calculations of Sensitivity and SpecificityTermDefinitionClassification from tumor-normal variant comparisonClassification from tumor-only filtration algorithmA = True somatic call (TSC)Variants classified as somatic by tumor-only filtration algorithm and known to be true somatic variants from tumor-normal comparisonSomaticSomaticB = True germline call (TGC)Variants classified as germline by tumor-only filtration algorithm and known to be true germline variants from tumor-normal comparisonGermlineGermlineC = False somatic call (FSC)Variants classified as somatic by the tumor-only filtration algorithm but known to be true germline variants from tumor-normal comparisonGermlineSomaticD = False germline call (FGC)Variants classified as germline by the tumor-only filtration algorithm but known to be true somatic variants from tumor-normal comparisonSomaticGermlineTermDefinitionFormulaSensitivityProportion of true somatic variants (A) correctly identified by the tumor-only filtration algorithm as somatic out of all true somatic variants (A + D)A/A + DSpecificityProportion of true germline variants (B) correctly identified as germline by the tumor-only filtration algorithm out of all true germline variants (B + C)B/B + CPositive predictive value (PPV)Proportion of true somatic variants (A) correctly identified by the tumor-only filtration algorithm as somatic out of all variants called somatic by the algorithm (A + C)A/A + CNegative predictive value (NPV)Proportion of true germline variants (B) correctly identified as germline by the tumor-only filtration algorithm out of all variants called germline (B + D)B/B + D Open table in a new tab To model the ideal tumor-only variant detection algorithm, the accuracy of variant classification was assessed by databases within the tumor-only filtration algorithm via recursive partitioning using the party package (https://cran.r-project.org/web/packages/party/index.html, last accessed February 23, 2017) in the R coding environment. Recursive partitioning considered the following independent dichotomous variables: PVD MAF (≥1% or <1%), present in COSMIC (two or more occurrences) or not present in COSMIC (one or no occurrences), present or absent in the Clinical Interpretations of Variants in Cancer (CIViC) database,23Griffith M. Spies N.C. Krysiak K. McMichael J.F. Coffman A.C. Danos A.M. et al.CIViC is a community knowledgebase for expert crowdsourcing the clinical interpretation of variants in cancer.Nat Genet. 2017; 49: 170-174Crossref PubMed Scopus (312) Google Scholar reported as benign/likely benign or as pathogenic/likely pathogenic in ClinVar, present or not in HGMD, predicted as benign or predicted as damaging in missense mutation protein effect prediction algorithms (SIFT24Kumar P. Henikoff S. Ng P.C. Predicting the effects of coding non-synonymous variants on protein function using the SIFT algorithm.Nat Protoc. 2009; 4: 1073-1081Crossref PubMed Scopus (5027) Google Scholar, 25Ng P.C. Henikoff S. SIFT: predicting amino acid changes that affect protein function.Nucleic Acids Res. 2003; 31: 3812-3814Crossref PubMed Scopus (4109) Google Scholar; PolyPhen2 HumDiv or HumVar26Adzhubei I.A. Schmidt S. Peshkin L. Ramensky V.E. Gerasimova A. Bork P. Kondrashov A.S. Sunyaev S.R. A method and server for predicting damaging missense mutations.Nat Methods. 2010; 7: 248-249Crossref PubMed Scopus (9354) Google Scholar; Mutation Taster27Schwarz J.M. Rödelsperger C. Schuelke M. Seelow D. MutationTaster evaluates disease-causing potential of sequence alterations.Nat Methods. 2010; 7: 575-576Crossref PubMed Scopus (2138) Google Scholar; Mutation Assessor28Reva B. Antipin Y. Sander C. Predicting the functional impact of protein mutations: application to cancer genomics.Nucleic Acids Res. 2011; 39: e118Crossref PubMed Scopus (1286) Google Scholar; LRT29Chun S. Fay J.C. Identification of deleterious mutations within three human genomes.Genome Res. 2009; 19: 1553-1561Crossref PubMed Scopus (684) Google Scholar; PROVEAN30Choi Y. Chan A.P. PROVEAN web server: a tool to predict the functional effect of amino acid substitutions and indels.Bioinformatics. 2015; 31: 2745-2747Crossref PubMed Scopus (1506) Google Scholar), and present or not in our internal laboratory CR-MVL. The performance of full (all variables) and minimal (three selected variables) filtration algorithms was evaluated to determine the most optimal approach for building tumor-only classification filtration approaches. To evaluate the tumor-only variant filtration algorithm, an existing data set of 104,784 candidate variants from 1120 matched tumor FFPE and normal PBL sample DNA pairs was used and tested by a targeted 48-gene NGS panel (TSACP12Stockley T.L. Oza A.M. Berman H.K. Leighl N.B. Knox J.J. Shepherd F.A. Chen E.X. Krzyzanowska M.K. Dhani N. Joshua A.M. Tsao M.-S. Serra S. Clarke B. Roehrl M.H. Zhang T. Sukhai M.A. Califaretti N. Trinkaus M. Shaw P. van der Kwast T. Wang L. Virtanen C. Kim R.H. Razak A.R.A. Hansen A.R. Yu C. Pugh T.J. Kamel-Reid S. Siu L.L. Bedard P.L. Molecular profiling of advanced solid tumors and patient outcomes with genotype-matched clinical trials: the Princess Margaret IMPACT/COMPACT trial.Genome Med. 2016; 8: 109Crossref PubMed Scopus (156) Google Scholar) (Figure 1). Variants were classified as true somatic or true germline to evaluate the sensitivity and specificity of the tumor-only variant filtration algorithm for identification of somatic variants from tumor-only testing. After candidate variants with low VAF and low coverage were excluded, 27,522 variants remained, of which 3764 met the retention criteria of nonsynonymous variants in regions of interest (exons or first 2 bp of introns flanking exons). The tumor-only variant filtration algorithm classified 1892 variants as somatic and 1802 as germline. The 1892 variants classified as somatic contained 100% of the true somatic variants (1803 variants; TSC) known from the tumor-normal (FFPE-PBL) variant comparison, but overcalled true germline variants as somatic, with 4.7% of true germline variants (89 of 1892; FSC) misclassified as somatic. Conversely, of 1961 true germline variants, 95.5% (1872 of 1961; TGC) were correctly identified as germline by the tumor-only variant filtration algorithm, with the remaining 4.5% (89 of 1961; FSC) of true germline variants misclassified as somatic. Overall, the tumor-only variant filtration algorithm provided a sensitivity for c