Predicting the functional impact of KCNQ1 variants with artificial neural networks

Abstract

Recent advances in experimental and computational protein structure determination have provided access to high-quality structures for most human proteins and mutants thereof. However, linking changes in structure in protein mutants to functional impact remains an active area of method development. If successful, such methods can ultimately assist physicians in taking appropriate treatment decisions. This work presents three artificial neural network (ANN)-based predictive models that classify four key functional parameters of KCNQ1 variants as normal or dysfunctional using PSSM-based evolutionary and/or biophysical descriptors. Recent advances in predicting protein structure and variant properties with artificial intelligence (AI) rely heavily on the availability of evolutionary features and thus fail to directly assess the biophysical underpinnings of a change in structure and/or function. The central goal of this work was to develop an ANN model based on structure and physiochemical properties of KCNQ1 potassium channels that performs comparably or better than algorithms using only on PSSM-based evolutionary features. These biophysical features highlight the structure-function relationships that govern protein stability, function, and regulation. The input sensitivity algorithm incorporates the roles of hydrophobicity, polarizability, and functional densities on key functional parameters of the KCNQ1 channel. Inclusion of the biophysical features outperforms exclusive use of PSSM-based evolutionary features in predicting activation voltage dependence and deactivation time. As AI is increasing applied to problems in biology, biophysical understanding will be critical with respect to ‘explainable AI’, i.e., understanding the relation of sequence, structure, and function of proteins. Our model is available at www.kcnq1predict.org.

Evolutionary coupling analysis guides identification of mistrafficking-sensitive variants in cardiac K+ channels: Validation with hERG

“…For example, application of computational prediction methods (including PolyPhen2, and PROVEAN) to LQTS genes has been shown to be moderately successful for predicting pathogenic KCNQ1 (LQTS type 1) and hERG (KCNH2; LQTS type 2) variants but less so for cardiac Na + channel variants (SCN5A; LQTS type 3) ( Leong et al, 2015 ) where LQTS (type 3) pathogenicity is predominantly associated with a gain of function (GOF) phenotype, with gating deficiencies (perturbed inactivation) in otherwise folding-competent pathogenic SCN5A variants ( Giudicessi et al, 2018 ). New channel-specific computational tools [e.g., ( Heyne et al, 2020 ; Zhang et al, 2020 ; Phul et al, 2022 )] may improve variant phenotype predictability, although experimental phenotyping is still required to improve confidence for clinical decision making. While EC analysis may be informative in structural analysis of Na + channels it is not expected to be useful in itself for predicting SCN5A LQTS type 3 variant phenotypes.…”

MLe-KCNQ2: An Artificial Intelligence Model for the Prognosis of Missense KCNQ2 Gene Variants

“…We suggest further work on the dataset for missense variants of KCNQ2 by incorporating unseen variants from the gnomAD database or recently reported studies. Designing new features for variant characterization, such as the change in the number of hydrogen donor or acceptor sites, would improve classification metrics as proposed in [ 58 ]. Advances in protein structure prediction (e.g., AlphaFold2) as well as cryo-EM technologies could lead to the design of more complex 3D features that could lead to a breakthrough in the prediction of variant pathogenicity.…”

ImprovedKCNQ2gene missense variant interpretation with artificial intelligence

“…This work pursued two main objectives: (1) developing a ML model to improve accuracy for predicting the pathogenicity of KCNQ2 missense variants by combining biological-evolutive information, physico-chemical changes and structural features; and (2) creating a framework that can be extended to the prediction of the pathogenicity of variants occurring in other genes. Although there was currently no specific ML based software for KCNQ2 , similar approaches have been applied to KCNQ1 , another member of the KCNQ family [Phul et al ., 2022; Draelos et al ., 2022; Li et al ., 2017].…”

“…While tools like FATHMM, CONDEL, PolyPhen2, MutPred2, SIFT or PROVEAN were trained using information from genome-wide genetic variation data [Pejaver et al ., 2020; Shihab et al ., 2013; Choi et al ., 2012; González-Pérez & López-Bigas, 2011; Adzhubei et al ., 2010; Ng & Henikoff, 2001]; MLe-KCNQ2 was specifically trained with KCNQ2 missense variants. This reflects that development of genome-wide tools was based on heterogeneous datasets including a wide range of proteins with diverse functions and associated diseases [Phul et al ., 2022]. As a direct consequence, their prediction accuracy can vary between genes [Richards et al ., 2015].…”

“…As a consequence, we suggest further work on the dataset for missense variants of KCNQ2 by incorporating unseen variants from gnomAD database (https://gnomad.broadinstitute.org/) or recently reported studies. Designing new features for variant characterization such as change in number of hydrogen donor sites or change in number of hydrogen acceptor sites would improve classification metrics as some works remark [Phul et al ., 2022]. Advances in protein structure prediction (e.g., AlphaFold2) as well as cryo-EM technologies could lead in the design of more complex 3D-features that could make a breakthrough in the prediction of variant pathogenicity.…”