B-324 Pgxtools: a comprehensive platform with multiple bioinformatics pipelines for next generation sequencing based pharmacogenomics workflows

Abstract

Pharmacogenomics (PGx) testing supports personalized drug treatment based on a patient’s genetic profile, with the goal to optimize drug selection/dosing, prevent adverse drug reactions and improve patient safety. Several single gene-drug-pair tests have become part of the standard-of-care. Multigene-panel-based pre-emptive PGx tests have also been implemented in some large centers. Compared with genotyping methods, next-generation-sequencing (NGS) technology offers advantages of covering rare variants in diverse populations, detecting new pathological variants, and allowing reanalysis of data when new evidence is discovered. However, clinical diagnostic laboratories are facing barriers in sequencing data analysis and variant interpretation with clinical decision support. In this study, we aim to develop an automated cloud-native platform for both long-read and short-read NGS workflows, integrated with the Clinical Pharmacogenetics Implementation Consortium (CPIC) database to provide clinical decisions. Bioinformatics platform design: an end-to-end pharmacogenomics platform, named Pgxtools, was developed to streamline the secondary and tertiary NGS data analysis for variant calling, star allele diplotyping and PGx recommendations aligning with CPIC guidelines. Through a password protected user interface, operators without bioinformatics training can easily upload NGS data and generate PGx reports. Cloud-based architecture allows both developers and end-users to access the platform remotely. Long-read NGS data: 16 genomic DNAs (12 Coriell GeT-RM cell lines and 4 patients) were sequenced in-house, using Twist Alliance Long-Read PGx Panel with Twist Long Read Library Preparation and Standard Hyb v2 Enrichment protocol, followed by HiFi long-read sequencing with PacBio Revio analyzer. BAM and VCF files were generated with SMRT® Link Target Enrichment Workflow. Reagents were purchased from Twist Bioscience and PacBio. Short-read whole genome sequencing (WGS) data: WGS alignments (BAM/CRAM) of 8 GeT-RM samples were downloaded from the 1000 Genomes Project (G1K) dataset. Comparison of PGx results: both long-read and short-read sequence data were analyzed by Pgxtools and compared with GeT-RM reference results and/or genotyping results by other methods. Three pipelines were created for PGx workflows under the Pgxtools platform: (1) for short-reads, Bcftools (for all other genes) plus HiFiHLA (for HLA genes) and BCyrius (for CYP2D6), (2) for long reads of all genes, Pb-StarPhase, and (3) for long reads, Bcftools plus HiFiHLA and Pb-StarPhase (for CYP2D6 only). Among 14 common PGx genes, comparison of the three pipelines yielded high concordance. Each pipeline has its unique features to accommodate the source of NGS data, and Bcftools computed faster than Pb-StarPhase. The results of the long-reads by in-house sequencing and short-reads from public datasets analyzed by Pgxtools showed very high concordance with the community consensus in all genes including CYP2D6 and HLA genes. PGx reports generated by Pgxtools for 14 common PGx genes in all 16 samples have been verified to be consistent with CPIC guidelines. Pgxtools is a bioinformatics platform created to support pre-emptive PGx testing and enable repurposing of existing WGS or whole exome sequencing data for PGx reporting. The genotyping results are coupled with dosing guidelines in automated, reproducible pipelines, facilitating rapid generation of clinically actionable PGx recommendations for gene-drug pairs and multigene-drug pairs.