How good are pathogenicity predictors in detecting benign variants?

Vihinen, M.; Niroula, A.

doi:10.1101/408153

Home/Archive/bioRxiv

bioRxiv · 10.1101/408153

How good are pathogenicity predictors in detecting benign variants?

Vihinen, M.·Niroula, A.

View original source ↗Export metadata ↓

Abstract

Computational tools are widely used for interpreting variants detected in sequencing projects. The choice of computational tools is critical for reliable variant impact interpretation for precision medicine and should be based on systematic performance assessment. The performance of the methods varies widely in different performance assessments, for example due to the sizes of test datasets. To address this issue, we obtained 63,160 common amino acid substitutions (allele frequency [≥]1% and <25%) from the Exome Aggregation Consortium (ExAC) database, which contains variants from 60,706 genomes or exomes. We evaluated the specificity, the capability to detect benign variants, for 10 variant interpretation tools. In addition to overall specificity of the tools, we tested their performance for variants in six geographical populations. PON-P2 had the best performance (95.5%) followed by FATHMM (86.4%) and VEST (83.5%). While these tools had excellent performance, the poorest method predicted more than one third of the benign variants to be disease-causing. The results allow choosing reliable methods for benign variant interpretation, for both research and clinical purposes, as well as provide a benchmark for method developers.\n\nAUTHOR SUMMARYIn precision/personalized medicine of many conditions it is essential to investigate individuals genome. Interpretation of the observed variation (mutation) sets is feasible only with computational approaches. We assessed the performance of variant pathogenicity/tolerance prediction programs on benign variants. Variants were obtained from high-quality ExAC database and selected to have minor allele frequency between 1 and 25%. We obtained 63,160 such cases and investigated 10 widely used predictors. Specificities of the methods showed large differences, from 0.64 to 0.96%, thus users of these methods have to be careful when choosing the one(s) they will use. We investigated further the performances on different populations, allele frequencies, separately for males and females, chromosome wise and for population unique and non-unique variants. The ranking of the tools remained the same in all these scenarios, i.e. the best methods were the best irrespective on how the data is filtered and grouped. This is to our knowledge the first evaluation of method performance on benign variants.

Source connections

Published article ↗

Explore related subjects

bioinformatics Preprint

Keep this discovery

Explore connections, maps & timelines

BibTeX RIS

Vihinen, M., Niroula, A.. 2018-09-04. How good are pathogenicity predictors in detecting benign variants?. https://doi.org/10.1101/408153

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related preprints

bioRxiv2026

mFrag: Integrating External DNA Methylation References Improves Lung Cancer Early Detection by Fragmentomics

Fragmentomics analyzes features of circulating cell-free DNA fragments (e.g., fragment length or end motifs) derived from whole-genome sequencing of liquid biopsy samples (blood or other body fluids), enabling non-invasive early cancer detection. Here we present mFrag, a framework that integrates external methylation reference atlases with fragmentomic features for lung cancer detection. mFrag comprises two complementary methods: mFrag-Weight, which uses DNA methylation information to weight fragmentomic features, and mFrag-Net, which integrates fragmentomic features and external DNA methylation data through neural networks. We average the output probabilities of the two methods to obtain a final prediction score. The mFrag model was trained using 287 individuals at risk for lung cancer. In an independent case-control testing set comprising 385 non-cancer controls and 46 cancers (28 stage I, 12 stage II, and 6 stage III or IV), mFrag outperforms a DELFI-style fragmentomic approach, improving PR-AUC from 0.455 to 0.720 and sensitivity at 99% specificity from 0.043 to 0.478, with ROC-AUC from 0.871 to 0.908. For stage I lung cancer, mFrag raises PR-AUC from 0.314 to 0.558 and ROC-AUC from 0.855 to 0.894. Applied to end 4-mer motif features, mFrag likewise improved over an unweighted baseline, with gains contributed by both mFrag-Weight and mFrag-Net. In summary, mFrag improves the accuracy of fragmentomics-based lung cancer detection through integrating external DNA methylation data and the complementarity of its two component models, without incurring any additional data-generating cost.

bioinformatics↗

bioRxiv2026

AOPTK: Open-Source Software for AI-Assisted Toxicological Literature Analysis

Motivation: Toxicologists today face a challenge: although vast amounts of data describing how chemicals affect human health are available across scientific publications and regulatory reports, much of this information remains underutilized because extracting and interpreting evidence across thousands of studies is impractical to perform manually. Recent advances in artificial intelligence (AI), particularly in large language models (LLMs), provide new opportunities to automate toxicological literature analysis. Results: Herein, we present the Adverse Outcome Pathway Toolkit (AOPTK), an open-source Python package for AI-enhanced analysis of toxicological and biomedical literature. AOPTK supports (i) automated literature retrieval from databases such as PubMed Central and Europe PMC, (ii) direct processing of full-text articles in PDF format, and (iii) LLM-assisted analysis tools to identify chemicals and extract their relationships to biological effects under user-defined conditions. We demonstrate its utilities through a case-study identifying potential non-genotoxic carcinogens based on their ability to inhibit gap junction intercellular communication (GJIC). The LLM-enhanced analysis achieves 81% agreement with an expert-curated inhibitor list and 95% accuracy in predicting inhibition versus no inhibition when analysing the same subset of 86 publications. Automated retrieval and analysis of 1 124 publications from PubMed Central further demonstrated the feasibility of applying AOPTK to large-scale literature analyses. Availability: The package is developed open source on GitHub, disseminated through PyPI and Bioconda and releases are archived on Zenodo. Individual tools are available in the Galaxy platform for accessible data analysis with user-friendly, findable, accessible, interoperable, and reusable (FAIR) and end-to-end workflows provided via WorkflowHub. The tools are available on usegalaxy.eu and usegalaxy.cz and are developed open source on GitHub.

bioinformatics↗

bioRxiv2026

MELD: Mutation Effect Learning through Differential embeddings

We introduce MELD (Mutation Effect Learning through Differential Embeddings), a compact model for predicting the functional consequences of single nucleotide variants (SNVs) by training directly on the wild-type (WT) versus mutant differential. Interpreting SNVs is central to clinical genomics, yet experimental characterization cannot scale to millions of observed variants. Leading sequence-to-function models (Enformer, Borzoi, AlphaGenome) are trained on absolute genomic tracks and score variants post hoc through in-silico mutagenesis (ISM), so the WT-mutant difference never enters the training loss. MELD instead places the logit difference {Delta}z = zmut - zwt inside the supervised objective. It combines a shared CNN-Transformer encoder, a mutant-specific mutation-mask boost, multiscale mutation-centered pooling, and a learned adaptive amplification factor . On ClinVar (1.57M variants), MELD (1.9M parameters) reached an AUROC of 0.986 versus 0.800 for a same-data, same-backbone ISM baseline, with ablations confirming the contribution of each component. On a shared benchmark, it outperformed frozen foundation models with 3-1,300x more parameters (AUROC 0.967 vs. 0.924 for AlphaGenome and 0.936 for Nucleotide Transformer). With only 50, 000 training variants, it exceeded the ISM baseline trained on all 1.48M (AUROC 0.903 vs. 0.796) variants. On BRCA1 saturation genome editing, frozen MELD achieved a Spearman {rho} of 0.42, above Borzoi (0.40), and ClinVar transfer raised it to 0.45. These results demonstrate that direct differential supervision is a compact, data-efficient alternative to post-hoc ISM scoring with substantially larger pretrained models.

bioinformatics↗