Written informed consent to participate in the study was obtained from patients with solid tumors treated at the Indiana University Precision Genomics Clinic (Indianapolis, IN) from September 2017 to March 2020. Patients consented to provide whole blood samples for genomic analyses, which consisted of (1) pharmacogenetics genotyping performed at the Indiana University Pharmacogenomics Laboratory and (2) WGS performed at NantOmics or WES performed at Ashion Analytics. The study enrolled a total of 100 patients with WGS and 164 patients with WES. The study protocol and the parent Indiana University Total Cancer Care Protocol were approved by the Indiana University Institutional Review Board. Patient demographic and clinical data, including age, sex, race/ethnicity, and date and type of first cancer diagnosis, were obtained from a Total Cancer Care Protocol study database.

We have previously described our panel-based genotyping method and CYP2D6 copy number testing in detail (17, 19). To summarize, genotyping was performed at the Clinical Laboratory Improvement Amendments (CLIA)-certified, College of American Pathologists (CAP)-accredited Indiana University Pharmacogenomics Laboratory using a validated, custom-designed OpenArray^® platform (ThermoFisher, Waltham, MA). The genes included on the genotyping platform were CYP2B6, CYP2C8, CYP2C9, CYP2C19, CYP2D6 (including copy number), CYP3A4, CYP3A5, CYP4F2, DPYD, G6PD, SLCO1B1, TPMT, and VKORC1. The complete list of the 45 variants on the genotyping platform are shown in Supplemental Table 1 . NUDT15 was added to the genotyping platform after genotyping was completed for the WGS cohort, necessitating orthogonal confirmation for any NUDT15 variant called by Aldy in the WGS cohort. Patients whose DNA sample passed clinical laboratory quality control (all 161 patients in the WES cohort and 94/100 patients in the WGS cohort) were genotyped.

Orthogonal confirmation was performed for any variant in the 14 assessed pharmacogenes that was detected by Aldy, not included in the genotyping platform, and had actionable recommendations in current Clinical Pharmacogenetics Implementation Consortium (CPIC) guidelines. Detailed descriptions of these orthogonal confirmation methods have been previously published (17). To summarize, these variants were confirmed with PCR-based genotyping using commercial TaqMan reagents, when available, and a positive genomic control from a Coriell cell line with known genotypes from 1000 Genomes for TaqMan allelic discrimination. All other variants were confirmed using Sanger sequencing methods, with primers for each variant being designed using the National Center for Biotechnology Information Primer-Blast tool (https://www.ncbi.nlm.nih.gov/tools/primer-blast) and synthesized by Integrated DNA Technologies, Inc. The assay identification numbers of the TaqMan reagents and the sequences of the Sanger sequencing primers are provided in Supplemental Table 2 .

Whole blood samples were sent to the following CLIA-certified, CAP-accredited clinical laboratories for next-generation sequencing: NantOmics Laboratory (Culver City, CA) for WGS or Ashion Analytics Laboratory (Phoenix, AZ) for WES. Germline NGS was performed as part of each laboratory’s commercial paired normal/tumor sequencing workflow, available as GPS Cancer^™ from NantOmics and as Genomic Enabled Medicine Exome Testing (GEM ExTra) from Ashion Analytics. NGS methods for GPS Cancer^™ and GEM ExTra have been previously published (20, 21). Briefly, germline DNA was isolated from whole blood samples, and DNA sequencing libraries were prepared using a custom xGEN target capture kit (GEM ExTra) or a KAPA Hyper prep kit (GPS Cancer^™). Libraries were sequenced using Illumina sequencing platforms, and sequencing data were analyzed using proprietary bioinformatics workflows. Sequencing reads were aligned to the Genome Reference Consortium Human Build 37 (GRCh37)/Human Genome version 19 (hg19). Binary alignment map (BAM) and variant calling files (VCFs) were returned from each laboratory and transferred into the Precision Health Cloud (LifeOmic, Indianapolis, IN) for analysis.

The Aldy and Cyrius genotype extraction methods have been described in detail in previous publications (9, 13, 18). Aldy v3.3 and v4.4 were downloaded via pypi packages (links available on the Aldy Github: https://github.com/0xTCG/aldy), and the Cyrius v1.1.1 source code was downloaded from the Cyrius Github (https://github.com/Illumina/Cyrius/releases/tag/v1.1.1). All callers were operationalized in the LifeOmic Precision Health Cloud using JupyterLab notebooks. Read depth was analyzed from NantOmics WGS BAM files for all variants included in the genotyping platform using Samtools and outputted in the.csv file format for analysis. The same method was previously used to analyze read depth from Ashion WES BAM files (17). Aldy performed genotype extraction for the major pharmacogenes contained on the genotyping platform, which included 14 genes from WGS BAM files and 13 genes from WES BAM files (VKORC1 was not assessed using WES BAMs since the rs9923231 promoter variant was not covered). The “WXS” and “Illumina” sequencing profiles were used for Aldy genotype extraction from WES data and WGS data, respectively. All other default Aldy parameters were used. Aldy output files, which were outputted in the.txt file format for analysis, included the list of variants detected and the assembled diplotype call. Cyrius was used to perform genotype extraction for the four BAM files that had discordant CYP2D6 genotype calls with Aldy v3.3 or v4.4 (or both) relative to the genotyping reference standard.

The accuracy of Aldy calls was assessed at the variant and diplotype levels via comparison to clinical laboratory genotype results or results from orthogonal methods. Diplotypes called by Aldy with ambiguous phasing or with hybrid alleles calls containing CYP2D6 and CYP2D7 (herein referred to as “CYP2D6 hybrid alleles”) were excluded from the diplotype concordance analysis. Standard analytical parameters for diagnostic testing, including analytical sensitivity and specificity, intra- and inter-assay concordance, and overall accuracy, were assessed for each PGx-relevant variant in our cohort possessed by ≥1 study patient. Additional star alleles identified by Aldy were deemed “clinically actionable” if they had clinical function annotations other than “normal function” by the Pharmacogene Variation Consortium (PharmVar) and PharmGKB (22, 23).

For patients with additional clinically actionable star alleles identified by Aldy (that were able to be unambiguously phased), medication data was obtained since each patient’s respective date of first cancer diagnosis via query of the Indiana Health Information Exchange, a statewide electronic health record data repository, as previously described (19). Patient pharmacogene phenotypes were categorized based on current diplotype to phenotype translation tables curated by PharmGKB and CPIC (22). Patient medications were considered actionable if current CPIC guidelines provide “strong” or “moderate” recommendations to alter therapy with the medication (i.e., prescribe an alternative medication, adjust the dose, increase monitoring, or perform additional functional testing) based on each patient’s Aldy-predicted phenotype (relative to their phenotype as predicted by the genotyping reference standard) (24). Medications that were considered actionable for each patient are summarized in Supplemental Table 3 .

For comparison to the genotyping reference standard, Aldy calls were extracted from clinical NGS data in a cohort of 264 patients, which consisted of 100 patients with clinical WGS and 164 patients with clinical WES. The demographic and clinical characteristics of the overall study population are presented in Table 1 , with characteristics for the WGS cohort provided in Supplemental Table 4 . Demographic and clinical characteristics for the WES cohort have been previously published (17). Within the overall study population, the median age at first cancer diagnosis was 56 years old (interquartile range: 16). Our population was 53% male, 85% white race, and 99% non-Hispanic ethnicity. The most frequent primary cancer diagnoses in our population were breast (15%), colorectal (12%), pancreatic (11%), prostate (10%), and soft tissue sarcoma (7%).

Read depth analysis was performed to confirm that the chromosomal positions for the 45 variants included on our genotyping panel were sufficiently covered within our NGS data. The average read depth at each position for the WGS cohort is shown in Supplemental Table 5 , and the average read depth for the WES cohort has been previously published (17). Within the WGS cohort, all PGx-relevant loci had an average read depth above the minimum recommended WGS coverage threshold of 30 reads except for the G6PD “A” and c.202G>A “A-” alleles, which had an average read depth of 29 ± 15 (mean ± standard deviation) reads (25). Within the WES cohort, all PGx loci had an average read depth above 100 reads except for CYP3A4*22, which had a read depth of 44 ± 12 reads. These findings confirmed that all PGx loci were adequately covered in the NGS data to proceed with the analytical assessment of the Aldy method.

An analysis validating Aldy v3.3 genotyping calls within the WES cohort has been previously published (17). For 94 patients in the WGS cohort, Aldy v3.3 calls were compared to the genotyping reference standard at both the variant and diplotype levels, with results shown in Tables 2 , 3 , respectively. Aldy v3.3 calls were concordant with the reference standard for all assessed variants within CYP2C8, CYP2C9, CYP2C19, CYP3A5, CYP4F2, DPYD, G6PD, NUDT15, SLCO1B1, TPMT, and VKORC1, as reflected by point estimates of 100% for analytical sensitivity, analytical specificity, and overall accuracy. The following discordant Aldy v3.3 calls were observed: one CYP2B6*6 (c.516G>T) call; two CYP2D6*2 (c.886C>T) calls; two CYP2D6*2 (c.1457G>C) calls; and one CYP3A4*22 (c.522-191C>T) call. Interrogation of the WGS data revealed that five of these heterozygous variants had variant allele frequencies just outside of the 25-75% threshold used by Aldy v3.3 to classify variants as heterozygous in diploid genotype calls, which explains the discordant results. Aldy v3.3’s intra- and inter-assay concordance, defined as the ability to reproduce the same call for the same sequencing file within the same and subsequent runs, respectively, were confirmed to be 100% for all assessed variants. Stemming from the variant level results, five diplotype calls (WGS16 in CYP2B6, WGS5, WGS35, and WGS55 in CYP2D6, and WGS73 in CYP3A4) were discordant. Aldy v3.3 identified additional alleles not covered by the genotyping reference standard for 196 diplotype calls, which included 15 clinically actionable alleles within CYP2B6 (1), CYP2C19 (3), DPYD (8), SLCO1B1 (2), and NUDT15 (1); these additional clinically actionable alleles were confirmed by orthogonal genotyping methods. All other 1,005 diplotype calls were concordant across the 14 assessed genes.

Aldy v4.4 diplotype calls were extracted from WES and WGS sequencing files. For three of the WES files, Aldy v4.4 returned a warning that “the average coverage was too low” and did not produce output. Table 4 summarizes Aldy v4.4 diplotype calls compared to the genotyping reference standard for the remaining 161 patients in the WES cohort across the 13 assessed pharmacogenes. With the exception of 23 CYP2D6 diplotypes that were excluded from analysis due to containing copy number variation, all Aldy v4.4 calls were concordant with the reference standard except for one CYP3A4 diplotype call (WES55) in one patient in which WES read depth at the CYP3A4*22 locus was <30. Aldy v4.4 identified additional alleles not covered by the genotyping standard for 385 diplotypes, which included 18 clinically actionable star alleles within CYP2B6 (2), CYP2C19 (1), CYP2D6 (6), DPYD (5), G6PD (1), NUDT15 (1), and SLCO1B1 (2). All additional clinically actionable alleles were orthogonally confirmed during this analysis or during our previous WES validation (17).

As shown in Table 5 , Aldy v4.4 diplotype calls from WGS data were concordant across all genes except for one CYP2C9 call (WGS48) and two CYP2D6 calls (WGS53 and 55). The CYP2C9 discordant call is puzzling since Aldy v3.3 produced the correct call, and the WGS variant allele frequency was 71% (Aldy v4.4 called the variant homozygous). Notably, two (WGS5 and 35) of the three discordant CYP2D6 calls from Aldy v3.3 were corrected by v4.4. However, one additional CYP2D6 discordant call (WGS53) by Aldy v4.4 was identified, which consisted of a copy number discrepancy. Cyrius was run on the four BAMs (WGS5, WGS35, WGS53, and WGS55) with discordant CYP2D6 Aldy v3.3 or v4.4 (or both calls), generating an accurate genotype call in all four cases (results shown in Supplemental Table 6 ). Aldy v4.4 did correct the CYP2B6 call (WGS16) that was miscalled by v3.3. Aldy v4.4 identified 232 additional alleles not covered by targeted genotyping in the WGS cohort, including the same 15 clinically actionable alleles that were previously identified by v3.3. Differences between additional major star alleles identified by Aldv v3.3 and v4.4 are due to new updates to Aldy’s gene databases from PharmVar with Aldy v4.4. The intra- and inter-assay concordance of Aldy v4.4 for all assessed variants was confirmed to be 100% using both WES and WGS data. All diplotype calls from Aldy v4.4, Aldy v3.3, and the genotyping reference standard for the WES and WGS cohorts are provided in Supplemental File 1 .

Thirty patients (11.5% of the study population), consisting of 13 and 17 patients from the WGS and WES cohorts, respectively, had additional clinically actionable star alleles conclusively identified by Aldy (two subjects [WGS69 and WES33] had additional clinically actionable CYP2B6 alleles that were unable to be unambiguously phased). Genotype-predicted phenotypes for these patients, as predicted by Aldy results and by results from the genotyping reference standard, are shown in Table 6 ; in addition, Table 6 summarizes prescriptions for actionable medications (i.e., those with different CPIC recommendations based on Aldy-predicted and genotyping-predicted phenotypes) since each patients’ respective date of first cancer diagnosis. Twenty-one of the 30 patients had medication data available, which included 289 ± 460 (median ± interquartile range) unique prescriptions per patient during the 7.3 ± 7.5 year follow-up period, defined as each patient’s date of first cancer diagnosis until the end of the data collection period. Of these 21 patients, 5 (23.8%) were prescribed a medication which had a different CPIC recommendation based on their Aldy-predicted pharmacogene phenotype.