Outside Front Cover: The cover image is based on the Special Article BRCA1- and BRCA2-specific in silico tools for variant interpretation in the CAGI 5 ENIGMA challenge by Natàlia Padilla et al., https://doi.org/10.1002/humu.23802. Cover design: Selen Özkan and Xavier de la Cruz.

Outside Back Cover: The cover image is based on the Overview article “Reports from the fifth edition of CAGI: The Critical Assessment of Genome Interpretation” by Gaia Andreoletiet al., https://doi.org/10.1002/humu.23876. Cover design: Zhiqiang Hu.

Pathogenic genetic variants often primarily affect splicing. In 2018, CAGI proposed two splicing prediction challenges based on experimental perturbation assays: Vex-seq, assessing exon skipping, and MaPSy assessing splicing efficiency. Our modular modeling framework, MMSplice, performed among the best for both challenges.

Based on the hypothesis that common functional pathways explain comorbidity between diverse neurodevelopmental disorders, we developed an efficient and cost-effective amplicon-based multigene panel to assess the pathogenic role of genes involved in intellectual disability (ID) and autism spectrum disorder (ASD) comorbidity. Here, we present the genetic findings after applying this panel to 150 individuals with ID and/or ASD.

Human frataxin (FXN) is an iron binding protein involved in mitochondrial Fe-S clusters assembly, a process fundamental for the functional activity of mitochondrial proteins. The study of the missense variants, selected from COSMIC database, suggests that the effect of the mutations is localised to the neighbourhoods of the mutated residues without affecting the global protein fold. The variants show a decreased stability and a decreased functional activity. Defective function of frataxin may cause defects in mitochondria, leading to increased tumorigenesis.

Cartoon showing the conceptual framework of phenotype alteration prediction based on coarse-grained molecular dynamics simulation that allows for quantification of structural flexibility change on residue mutation in protein. An example case of protein AvBD2 Defensin from Chicken (NMR structures with PDB ID: 2LG5 (wild type); 2LG6 (mutant)) shows how flexible protein becomes rigid on single mutation from Lysine to Alanine.

We developed a computational model that integrates general evolutionary and physiochemical features with contextual gene-specific features to predict the impact of genetic variants. Through blind predictions of residual enzymatic activity in human acid α-glucosidase (GAA) mutants, we demonstrate that gene-specific features can provide clues for investigating origin of variant pathogenicity, particularly for variants that are poorly predicted by existing methods.

Variation in BRCA1 and BRCA2 can greatly increase the risk of breast, ovarian and other cancers. Growing awareness of this fact is leading to an increased rate of genetic testing, which in turn has led to the discovery of thousands of Variants of Uncertain Significance (VUS). The rate of VUS discovery has outstripped the rate of variant interpretation, which has led for a growing need for accurate, high-throughput methods for variant analysis. In the CAG5 ENIGMA challenge, participants predicted the clinical significance of 326 variants, for which expert interpretations had been completed but not published. Six teams submitted blind predictions on these variants, with fourteen methods collectively. The best performance was achieved by the LEAP methods by Color Genomics, which successfully leveraged private data including an HGMD subscription and an internal library of genetic testing results. This emphasizes that there are still private data that could inform variant prediction. To the extent that such data can be made publicly available, science will benefit, and genetic testing patients by extension.

Novel machine learning models predict the probability of pathogenicity for missense variants. They are further interpreted to identify the most contributing features (and the most probable molecular mechanisms) for each variant predicted to be pathogenic. Finally, protein structural modeling of such variations validate the hypothesized molecular mechanisms.

Disruptive BRCA1 and BRCA2 germline variants increase the risk of hereditary breast and ovarian cancers. We present two families of in silico predictors (multiple linear regression and neural network) designed to identify them, and the validation of these tools in the fifth Critical Assessment of Genome Interpretation-ENIGMA challenge. Our tools generally outperform standard predictors, as shown in the heatmap: Diagonal and off-diagonal elements correspond to successful and failed predictions, respectively.