B-126 Comparison of Liftover Tools for Conversion to Genome Reference Consortium Human Build 38 from Human Genome 19

Abstract

Abstract Background Genome Reference Consortium Human Build 38 (GRCh38) was released with improvements such as accuracy and completeness over human genome 19 (hg19). However, GRCh38 has not been widely adopted in clinical laboratories because re-alignment is time-consuming and computationally-expensive. To address this issue, faster and more convenient liftover tools have been developed to convert genome coordinates between assemblies. The aim of this study was to compare the performance of liftover tools for conversion to GRCh38 from hg19. Methods Two variant call format (VCF) files aligned to hg19 (No. of clinical variants = 1 570 644) and GRCh38 were downloaded from ClinVar (clinvar_20221217.vcf.gz). The liftover tools such as CrossMap, NCBIRemap, and UCSCliftOver were used to convert genome coordinates from hg19 to GRCh38. We investigated the conversion failure of clinical variants and evaluated classification of variants provided by multiple submitters with assertion criteria and evidence according to ClinVar review status, to ensure the authenticity of the variants. Results The conversion rates of CrossMap, NCBIRemap, and UCSCliftOver were 99.64% (1 567 848/1 570 644), 99.52% (1 565 963/1 570 644) and 99.81% (1 570 560/1 570 644), respectively. The variants which failed conversion were identified in ABCC9 (n = 2), ACACA (n = 1), CNOT3 (n = 3), CYFIP1 (n = 1), DGAT1 (n = 9), EMG1 (n = 1), GAS2L2 (n = 5), GDF2 (n = 38), GPR179 (n = 96), HNF1B (n = 132), HYOU1 (n = 2), MBOAT7 (n = 5), MYO19 (n = 7), NSDHL (n = 15), ODAD3 (n = 1), ORAI1 (n = 2), PCGF2 (n = 3), PQBP1 (n = 19), PRPF31 (n = 56), PUF60 (n = 9), RBP3 (n = 52), RPS17 (n = 1), SARM1 (n = 1), SCARF2 (n = 1), SLC35A2 (n = 27), SLC39A4 (n = 25), SLC52A2 (n = 86), SURF1 (n = 65), TAF15 (n = 8), TFE3 (n = 2), TSEN34 (n = 15), WDR45 (n = 56), ZNHIT3 (n = 1) genes. The numbers of pathogenic or likely pathogenic variants among variants which failed conversion by CrossMap, NCBIRemap, and UCSCliftOver were 67, 115, and 0, respectively. Classification of converted variants from GRCh37 to GRCh38 remained unchanged. Conclusions This study demonstrated that UCSCliftOver generates most reliable results. This study suggests that liftover tools might be one of the practical alternatives for genome conversion in case that re-alignment approaches were not possible.