A comparison on predicting functional impact of genomic variants

Abstract

Single-nucleotide polymorphism (SNPs) may cause the diverse functional impact on RNA or protein changing genotype and phenotype, which may lead to common or complex diseases like cancers. Accurate prediction of the functional impact of SNPs is crucial to discover the ‘influential’ (deleterious, pathogenic, disease-causing, and predisposing) variants from massive background polymorphisms in the human genome. Increasing computational methods have been developed to predict the functional impact of variants. However, predictive performances of these computational methods on massive genomic variants are still unclear. In this regard, we systematically evaluated 14 important computational methods including specific methods for one type of variant and general methods for multiple types of variants from several aspects; none of these methods achieved excellent (AUC ≥ 0.9) performance in both data sets. CADD and REVEL achieved excellent performance on multiple types of variants and missense variants, respectively. This comparison aims to assist researchers and clinicians to select appropriate methods or develop better predictive methods.

A Multi-Layered Computational Structural Genomics Approach Enhances Domain-Specific Interpretation of Kleefstra Syndrome Variants in EHMT1

“…Commonly used 2D sequence-based variant calling methods include SNP&Go (14), PROVEAN (15), PolyPhen2 (16), Rhapsody (17), CADD (18), and REVEL (19). Combined annotators, such as CADD and REVEL, show better performance (20). However, we recognize that protein function is not solely determined by its chemical composition but also by its molecular structure’s spatial arrangement and dynamic nature.…”

Advancing Antibiotic Resistance Classification with Deep Learning Using Protein Sequence and Structure

“…Finally, ARGGNN achieves better results than the DL-based methods in the second setting ( > 50% id) while being less performant than both BLAST and DIAMOND. To improve the results even further, we can use the ensembling method of [32] and incorporate BLAST results when the identity is very high. Table 1 illustrates the superiority of the ESM + GraphSAGE model over the previous ones once more, with the Macro F1 score improving from 59.7% to 63.78%.…”

“…(5) ARG-SHINE [32] is an ensemble method that combines the output of BLAST along with a convolutional neural network applied to sequences, and information about the family, domain and motif of proteins. We compare only against their convolutional neural network component since it is the model's dominant component, while the ensemble is based mainly on BLAST results, which can be used with all other models to improve their performance.…”

“…The model was subsequently trained on the newly created COALA dataset where it outperformed alignment-based methods. Another recent method, ARG-SHINE [33], integrates multiple techniques in an ensemble. More specifically, it aggregates the results of a convolutional neural network applied to the sequences, a protein functional-based logistic regression model, and a BLAST-based model.…”

Contribution of rare coding variants to complex trait heritability

“…Despite efforts to functionally characterize RVs, the biological impact of roughly 400 rare, putatively disruptive mutations carried by each individual remains largely unknown 2 . Classification of RVs is challenging, and current algorithms do not always correctly predict their pathogenic characteristics 3,4 . Indeed, existing tools to classify RVs have typically been trained on conditions of Mendelian inheritance [5][6][7] , while most human phenotypes are complex and non-Mendelian in nature.…”