Article Figures & Data
Figures
- Figure 1. Overview of study workflow for SV prioritization benchmarking.
This workflow illustrates the evaluation process for eight SV prioritization tools, categorized into knowledge-driven and data-driven approaches. These tools were benchmarked across seven independent and curated datasets using three main criteria: (1) accuracy in pathogenicity prediction, (2) robustness in diverse genomic and biological contexts, and (3) usability, focusing on user accessibility and computational performance.
- Figure 2. Comparative performance of eight SV prioritization approaches.
(A) Correlation analysis between positive (pathogenic) and negative (benign) variant sets across the eight approaches, indicating the differentiation ability of each tool. (B) Distribution of pathogenicity scores for positive and negative sets, showing score separation across the tools. (C) Performance summary across all germline variants from ClinVar, measured by area under the curve.
- Figure 3. Performance of SV prioritization tools across genomic contexts.
SV type, length, and gene content: (A) Distribution of pathogenicity scores for deletion and duplication sets, illustrating score separation across tools by SV type. (B) Performance of each tool in deletions and duplications among germline ClinVar variants, evaluated by area under the curve (AUC). (C) Length distributions of deletions and duplications within the dataset. (D) AUCs performance over three lengths ranges (L1< 6*103, L2:6*103∼105, L3 >105) for deletions and duplications. (E) Distribution differences in protein-coding gene coverage between negative (benign) and positive (pathogenic) SV sets. (F) AUC comparison by gene context (disease-related, functional genes) for deletions and duplications, further categorized by SVs covering zero genes (No. genes = 0) and one or more genes (No. genes ≥ 1). AUC, area under the curve; SV, structural variant.
- Figure S1. Distribution and intersection analyses of disease-relevant and functionally relevant genes.
(A) Upset plot showing the distribution and overlap of disease-relevant gene sets across three databases: ACMG (American College of Medical Genetics), ClinGen (Clinical Genome Resource), and Orphanet. Vertical bars indicate the size of each intersection, while horizontal bars represent the total number of genes in each database. The largest group is unique to ClinGen (3,686 genes), with additional smaller overlaps across combinations of the three datasets. (B) Upset plot of functionally relevant genes across four experimental and phenotypic databases: cell culture, pTriplo (probability of triplosensitivity), MGI (Mouse Genome Informatics), and pHaplo (probability of haploinsufficiency). The highest number of unique genes is in cell culture (1,351 genes), with other notable intersections among database combinations. Vertical bars show intersection sizes, while horizontal bars display total gene counts in each dataset.
- Figure 4. Performance across approaches covering various biological mechanisms including noncoding SVs, long range SVs, somatic SVs, GWAS SV and eQTL SV.
AUC, area under the curve; SV, structural variant.
Tables
Softwarea Year Main language Assumption Classifier Training set Result URL AnnotSV (Version: 3.3.6) 2018 Tcl, Shell, Python ACMG ACMG Implementation of ACMG guideline. Annotation, scores https://lbgi.fr/AnnotSV/ CADD-SV (Version 1.1) 2022 Python, R Evolutionary fitness Random forest Randomly distributed SVs over the human autosomes, evolutionarily fixed chimpanzee and human-derived SVs. Scores https://cadd-sv.bihealth.org/ ClassifyCNV (Version 1.1.1) 2020 Python, Shell ACMG ACMG Implementation of ACMG guideline. Scores https://github.com/Genotek/ClassifyCNV dbCNV 2023 Perl, Shell Molecular functions Gradient boosted trees The ClinVar, dbVar, ClinGen, DGV, DECIPHER and gnomAD (accessed before January 2023) Classification https://github.com/lllllv-1/dbCNV StrVCTVRE (Version 1.7) 2022 Python Molecular functions on exons Random forest Rare SVs from ClinVar, gnomAD, and a recent great ape sequencing study. Scores https://strvctvre.berkeley.edu/ SVScore (Version 0.6) 2017 Perl, Shell SNPs-based CADD scores Derived from CADDb The precomputed SNP scores generated by CADD v1.3 Scores https://github.com/lganel/SVScore TADA (Version 1.0.2) 2022 Python, Shell Molecular functions related to long range interaction Random forest DECIPHER, Variants in the set published by Audano et al (2019), GnomAD, UK Biobank data set and DGV. Scores https://github.com/jakob-he/TADA/ XCNV 2022 R,Shell Molecular functions XGBoost The dbVar, ClinGen, DECIPHER v10.1, and DGV (accessed before January 2021). Scores https://github.com/kbvstmd/XCNV Benchmark dataset Positive set (number of positive variants) Negative set (number of negative variants) Germline SVsa from ClinVar and GnomAD “pathogenic” and “likely pathogenic” germline SVs from ClinVar (January. 2023–April. 2024) (N = 489). (1) ”benign” and “likely benign” germline SVs from ClinVar (January. 2023–April. 2024) (N = 93); (2) randomly select rare SVs with matched lengths with positive sets from gnomAD v4 (N = 396). Noncoding SVs and GnomAD Noncoding SVs from peer-reviewed publications (N = 6). Randomly select rare SVs with matched lengths with positive sets from gnomAD v4; No overlapped with protein coding genes listed at gencode v30lift37 (N = 6). Long range SVs and GnomAD SVs implicated in long-range interactions from peer-reviewed publications (N = 12). Randomly select rare SVs with matched lengths with positive sets from gnomAD v4 (N = 12). Somatic SVs Somatic SVs from COSMIC (v99) with recurrence >=2 and located on risk genes listed at oncoKB (N = 218). Randomly select somatic SVs from COSMIC (v99) with recurrence = 1 and no overlapped with risk genes listed at oncoKB (N = 238). Disease associated SVs from a GWAS and GnomAD Rare SVs which validated by replication listed at the peer-reviewed publication (N = 32). Randomly select rare SVs with matched lengths with positive sets from gnomAD v4 (N = 32). Functional relevant SVs from eQTL studies and GnomAD Rare SVs: aberrant gene expression is in multi tissues and the gene has dosage changed (N = 72). Randomly select rare SVs with matched lengths with positive sets from gnomAD v4 (N = 72). ↵a SVs including CNVs and deletions, duplications.
Software Knowledge driven Data driven AnnotSV ClassifyCNV CADD-SV dbCNV StrVCTVRE SVScore TADA XCNV Efficiency (second) 12 3 100 (20 cores) 38 19 5 16 24 Document quality Excellent Normal Good Normal Good Normal Good Good Installation cmd, conda, docker cmd conda cmd cmd, conda cmd cmd cmd Prerequisite dataset Yes Yes Yes No Yes Yes Yes Yes Genome build hg19, hg38 hg19, hg38 hg38 hg19 hg19, hg38 hg19 hg19 hg19 Input Bed, vcf Bed Bed Bed Bed vcf Bed Bed SV types Deletion, insertion, duplication, inversion, breakend record CNV Deletion, insertion and duplication CNV Deletion, duplication All types CNV CNV Online webserver Yes No Yes No Yes No No Yes
Table S3. Performances over all approaches.
In this Issue
Related Articles
Cited By...
- No citing articles found.