In-silico phenotype prediction by normal mode variant analysis in TUBB4A-related disease

TUBB4A-associated disorder is a rare condition affecting the central nervous system. It displays a wide phenotypic spectrum, ranging from isolated late-onset torsion dystonia to a severe early-onset disease with developmental delay, neurological deficits, and atrophy of the basal ganglia and cerebellum, therefore complicating variant interpretation and phenotype prediction in patients carrying TUBB4A variants. We applied entropy-based normal mode analysis (NMA) to investigate genotype–phenotype correlations in TUBB4A-releated disease and to develop an in-silico approach to assist in variant interpretation and phenotype prediction in this disorder. Variants included in our analysis were those reported prior to the conclusion of data collection for this study in October 2019. All TUBB4A pathogenic missense variants reported in ClinVar and Pubmed, for which associated clinical information was available, and all benign/likely benign TUBB4A missense variants reported in ClinVar, were included in the analysis. Pathogenic variants were divided into five phenotypic subgroups. In-silico point mutagenesis in the wild-type modeled protein structure was performed for each variant. Wild-type and mutated structures were analyzed by coarse-grained NMA to quantify protein stability as entropy difference value (ΔG) for each variant. Pairwise ΔG differences between all variant pairs in each structural cluster were calculated and clustered into dendrograms. Our search yielded 41 TUBB4A pathogenic variants in 126 patients, divided into 11 partially overlapping structural clusters across the TUBB4A protein. ΔG-based cluster analysis of the NMA results revealed a continuum of genotype–phenotype correlation across each structural cluster, as well as in transition areas of partially overlapping structural clusters. Benign/likely benign variants were integrated into the genotype–phenotype continuum as expected and were clearly separated from pathogenic variants. We conclude that our results support the incorporation of the NMA-based approach used in this study in the interpretation of variant pathogenicity and phenotype prediction in TUBB4A-related disease. Moreover, our results suggest that NMA may be of value in variant interpretation in additional monogenic conditions.

Benchmarking Geometric Deep Learning for Cortical Segmentation and Neurodevelopmental Phenotype Prediction

The emerging field of geometric deep learning extends the application of convolutional neural networks to irregular domains such as graphs, meshes and surfaces. Several recent studies have explored the potential for using these techniques to analyse and segment the cortical surface. However, there has been no comprehensive comparison of these approaches to one another, nor to existing Euclidean methods, to date. This paper benchmarks a collection of geometric and traditional deep learning models on phenotype prediction and segmentation of sphericalised neonatal cortical surface data, from the publicly available Developing Human Connectome Project (dHCP). Tasks include prediction of postmenstrual age at scan, gestational age at birth and segmentation of the cortical surface into anatomical regions defined by the M-CRIB-S atlas. Performance was assessed not only in terms of model precision, but also in terms of network dependence on image registration, and model interpretation via occlusion. Networks were trained both on sphericalised and anatomical cortical meshes. Findings suggest that the utility of geometric deep learning over traditional deep learning is highly task-specific, which has implications for the design of future deep learning models on the cortical surface. The code, and instructions for data access, are available from https://github.com/Abdulah-Fawaz/Benchmarking-Surface-DL.

Improving Neural Networks for Genotype-Phenotype Prediction Using Published Summary Statistics

Phenotype prediction is a necessity in numerous applications in genetics. However, when the size of the individual-level data of the cohort of interest is small, statistical learning algorithms, from linear regression to neural networks, usually fail due to insufficient data. Fortunately, summary statistics from genome-wide association studies (GWAS) on other large cohorts are often publicly available. In this work, we propose a new regularization method, namely, main effect prior (MEP), for making use of GWAS summary statistics from external datasets. The main effect prior is generally applicable for machine learning algorithms, such as neural networks and linear regression. With simulation and real-world experiments, we show empirically that MEP improves the prediction performance on both homogeneous and heterogeneous datasets. Moreover, deep neural networks with MEP outperform standard baselines even when the training set is small.

Inbreeding Calculated with Runs of Homozygosity Suggests Chromosome-Specific Inbreeding Depression Regions in Line 1 Hereford

The goal of this study was to evaluate inbreeding in a closed beef cattle population and assess phenotype prediction accuracy using inbreeding information. Effects of inbreeding on average daily gain phenotype in the Line 1 Hereford cattle population were assessed in this study. Genomic data were used to calculate inbreeding based on runs of homozygosity (ROH), and pedigree information was used to calculate the probability of an allele being identical by descent. Prediction ability of phenotypes using inbreeding coefficients calculated based on pedigree information and runs of homozygosity over the whole genome was close to 0, even in the case of significant inbreeding coefficient effects. On the other hand, inbreeding calculated per individual chromosomes’ ROH yielded higher accuracies of prediction. Additionally, including only ROH from chromosomes with higher predicting ability further increased prediction accuracy. Phenotype prediction accuracy, inbreeding depression, and the effects of chromosome-specific ROHs varied widely across the genome. The results of this study suggest that inbreeding should be evaluated per individual regions of the genome. Moreover, mating schemes to avoid inbreeding depression should focus more on specific ROH with negative effects. Finally, using ROH as added information may increase prediction of the genetic merit of animals in a genomic selection program.

Controlling for background genetic effects using polygenic scores improves the power of genome-wide association studies

Ongoing increases in the size of human genotype and phenotype collections offer the promise of improved understanding of the genetics of complex diseases. In addition to the biological insights that can be gained from the nature of the variants that contribute to the genetic component of complex trait variability, these data bring forward the prospect of predicting complex traits and the risk of complex genetic diseases from genotype data. Here we show that advances in phenotype prediction can be applied to improve the power of genome-wide association studies. We demonstrate a simple and efficient method to model genetic background effects using polygenic scores derived from SNPs that are not on the same chromosome as the target SNP. Using simulated and real data we found that this can result in a substantial increase in the number of variants passing genome-wide significance thresholds. This increase in power to detect trait-associated variants also translates into an increase in the accuracy with which the resulting polygenic score predicts the phenotype from genotype data. Our results suggest that advances in methods for phenotype prediction can be exploited to improve the control of background genetic effects, leading to more accurate GWAS results and further improvements in phenotype prediction.

Controlling for Background Genetic Effects Using Polygenic Scores Improves the Power of Genome-wide Association Studies

Ongoing increases in the size of human genotype and phenotype collections offer the promise of improved understanding of the genetics of complex diseases. In addition to the biological insights that can be gained from the nature of the variants that contribute to the genetic component of complex trait variability, these data bring forward the prospect of predicting complex traits and the risk of complex genetic diseases from genotype data. Here we show that advances in phenotype prediction can be applied to improve the power of genome-wide association studies. We demonstrate a simple and efficient method to model genetic background effects using polygenic scores derived from SNPs that are not on the same chromosome as the target SNP. Using simulated and real data we found that this can result in a substantial increase in the number of variants passing genome-wide significance thresholds. This increase in power to detect trait-associated variants also translates into an increase in the accuracy with which the resulting polygenic score predicts the phenotype from genotype data. Our results suggest that advances in methods for phenotype prediction can be exploited to improve the control of background genetic effects, leading to more accurate GWAS results and further improvements in phenotype prediction.