Understanding structure-guided variant effect predictions using 3D convolutional neural networks.

Predicting pathogenicity of missense variants in molecular diagnostics remains a challenge despite the available wealth of data, such as evolutionary information, and the wealth of tools to integrate that data. We describe DeepRank-Mut, a configurable framework designed to extract and learn from physicochemically relevant features of amino acids surrounding missense variants in 3D space. For each variant, various atomic and residue-level features are extracted from its structural environment, including sequence conservation scores of the surrounding amino acids, and stored in multi-channel 3D voxel grids which are then used to train a 3D convolutional neural network (3D-CNN). The resultant model gives a probabilistic estimate of whether a given input variant is disease-causing or benign. We find that the performance of our 3D-CNN model, on independent test datasets, is comparable to other widely used resources which also combine sequence and structural features. Based on the 10-fold cross-validation experiments, we achieve an average accuracy of 0.77 on the independent test datasets. We discuss the contribution of the variant neighborhood in the model's predictive power, in addition to the impact of individual features on the model's performance. Two key features: evolutionary information of residues in the variant neighborhood and their solvent accessibilities were observed to influence the predictions. We also highlight how predictions are impacted by the underlying disease mechanisms of missense mutations and offer insights into understanding these to improve pathogenicity predictions. Our study presents aspects to take into consideration when adopting deep learning approaches for protein structure-guided pathogenicity predictions.

Breast cancer risks associated with missense variants in breast cancer susceptibility genes.

BACKGROUND: Protein truncating variants in ATM, BRCA1, BRCA2, CHEK2, and PALB2 are associated with increased breast cancer risk, but risks associated with missense variants in these genes are uncertain. METHODS: We analyzed data on 59,639 breast cancer cases and 53,165 controls from studies participating in the Breast Cancer Association Consortium BRIDGES project. We sampled training (80%) and validation (20%) sets to analyze rare missense variants in ATM (1146 training variants), BRCA1 (644), BRCA2 (1425), CHEK2 (325), and PALB2 (472). We evaluated breast cancer risks according to five in silico prediction-of-deleteriousness algorithms, functional protein domain, and frequency, using logistic regression models and also mixture models in which a subset of variants was assumed to be risk-associated. RESULTS: The most predictive in silico algorithms were Helix (BRCA1, BRCA2 and CHEK2) and CADD (ATM). Increased risks appeared restricted to functional protein domains for ATM (FAT and PIK domains) and BRCA1 (RING and BRCT domains). For ATM, BRCA1, and BRCA2, data were compatible with small subsets (approximately 7%, 2%, and 0.6%, respectively) of rare missense variants giving similar risk to those of protein truncating variants in the same gene. For CHEK2, data were more consistent with a large fraction (approximately 60%) of rare missense variants giving a lower risk (OR 1.75, 95% CI (1.47-2.08)) than CHEK2 protein truncating variants. There was little evidence for an association with risk for missense variants in PALB2. The best fitting models were well calibrated in the validation set. CONCLUSIONS: These results will inform risk prediction models and the selection of candidate variants for functional assays and could contribute to the clinical reporting of gene panel testing for breast cancer susceptibility.

Functional Analysis Identifies Damaging CHEK2 Missense Variants Associated with Increased Cancer Risk.

Heterozygous carriers of germline loss-of-function variants in the tumor suppressor gene checkpoint kinase 2 (CHEK2) are at an increased risk for developing breast and other cancers. While truncating variants in CHEK2 are known to be pathogenic, the interpretation of missense variants of uncertain significance (VUS) is challenging. Consequently, many VUS remain unclassified both functionally and clinically. Here we describe a mouse embryonic stem (mES) cell-based system to quantitatively determine the functional impact of 50 missense VUS in human CHEK2. By assessing the activity of human CHK2 to phosphorylate one of its main targets, Kap1, in Chek2 knockout mES cells, 31 missense VUS in CHEK2 were found to impair protein function to a similar extent as truncating variants, while 9 CHEK2 missense VUS resulted in intermediate functional defects. Mechanistically, most VUS impaired CHK2 kinase function by causing protein instability or by impairing activation through (auto)phosphorylation. Quantitative results showed that the degree of CHK2 kinase dysfunction correlates with an increased risk for breast cancer. Both damaging CHEK2 variants as a group [OR 2.23; 95% confidence interval (CI), 1.62-3.07; P < 0.0001] and intermediate variants (OR 1.63; 95% CI, 1.21-2.20; P = 0.0014) were associated with an increased breast cancer risk, while functional variants did not show this association (OR 1.13; 95% CI, 0.87-1.46; P = 0.378). Finally, a damaging VUS in CHEK2, c.486A>G/p.D162G, was also identified, which cosegregated with familial prostate cancer. Altogether, these functional assays efficiently and reliably identified VUS in CHEK2 that associate with cancer. SIGNIFICANCE: Quantitative assessment of the functional consequences of CHEK2 variants of uncertain significance identifies damaging variants associated with increased cancer risk, which may aid in the clinical management of patients and carriers.