Study Links Genetic Variants to Specific Disease Phenotypes

David Stein, Meltem Ece Kars, Baptiste Milisavljevic, Matthew Mort, Peter D. Stenson, Jean-Laurent Casanova, David N. Cooper, Bertrand Boisson, Peng Zhang, Avner Schlessinger, and Yuval Itan have expanded the utility of variant effect predictions through the development of phenotype-specific models. This work, conducted at institutions including the Icahn School of Medicine at Mount Sinai, Rockefeller University, Cardiff University, and INSERM U1163, focuses on refining predictions of variant effects. The research aims to improve understanding of the relationship between genetic variants and specific disease phenotypes, potentially advancing personalized medicine approaches and the identification of disease-causing mechanisms.

Expanding Variant Effect Prediction Utility Expanding the utility of variant effect predictions is a focus of current research, as evidenced by a manuscript accepted at Nature Communications on November 11, 2025. The work by Stein et al. aims to improve prediction methods, building upon existing tools like ClinVar, HGMD, and various algorithms for assessing missense variant effects (AlphaMissense, CADD v1.7). These improvements are critical for accurate variant interpretation and understanding disease mechanisms. Several studies highlight the importance of considering diverse data sources for accurate prediction. Research indicates that integrating protein language models, regulatory CNNs, and nucleotide-level scores enhances genome-wide variant prediction. Furthermore, disease-specific prediction tools—like VIPPID for primary immunodeficiency and DVPred for hearing loss—demonstrate that tailoring predictions to specific conditions improves accuracy and clinical relevance. Researchers are also exploring network-based methods and the “human gene connectome” to map short cuts for identifying morbid alleles. Approaches utilizing the Human Phenotype Ontology (HPO) and tools like PHENOstruct and HPOLabeler attempt to predict phenotypes associated with variants. These strategies, alongside hierarchical ensemble methods, are designed to improve the ability to link genetic variation to disease outcomes. Corresponding Authors and Affiliations The article identifies Yuval Itan and Avner Schlessinger as the Corresponding Authors for this research. Contact information is provided: yuval.itan@mssm.edu and avner.schlessinger@mssm.edu, respectively. Both authors have affiliations with the Department of Genetics and Genomic Sciences, and the AI Small Molecule Drug Discovery Center at Icahn School of Medicine at Mount Sinai in New York, NY, USA. This indicates a collaborative effort centered within these institutions. The author list includes affiliations spanning multiple institutions, demonstrating the scope of the study. These include Rockefeller University (New York, USA), Cardiff University (UK), and various research centers in France like INSERM U1163 and the Imagine Institute in Paris. Multiple researchers are listed, each with a specific institutional connection, highlighting the collaborative nature of this work and the diverse expertise involved. The research involved affiliations with several departments within Icahn School of Medicine at Mount Sinai. These include the Charles Bronfman Institute for Personalized Medicine, and departments focused on Artificial Intelligence and Human Health, and Pharmacological Sciences. This suggests an interdisciplinary approach integrating genetics, personalized medicine, and advanced computational methods to address the study’s goals. A general method applicable to the search for similarities in the amino acid sequence of two proteins. Reference to Related Genomic Studies This research builds upon a foundation of existing genomic studies and resources. References include work by Chen et al. (2024) mapping genomic mutational constraint, and Backman et al. (2021) detailing exome sequencing of 454,787 UK Biobank participants.

The Human Gene Mutation Database (HGMD®) by Stenson et al. (2020) is also cited, highlighting the importance of curated databases in variant analysis. These studies provide essential context for improving variant effect prediction. Several tools and approaches for variant analysis are referenced, demonstrating the field’s active development. Landrum et al. (2020) improved access to data through ClinVar, while Garcia et al. (2022) reviewed in silico tools for pathogenicity prediction. Resources like AlphaMissense, CADD v1.7, and guidelines for variant effect predictors (Livesey et al., 2024) are also noted, indicating a focus on developing and refining predictive algorithms. Disease-specific approaches to variant pathogenicity are also highlighted, with examples like VIPPID (Fang et al., 2022) for primary immunodeficiency, and DVPred (Bu et al., 2022) for hearing loss. Studies by Zhang et al. (2021) demonstrated improvements using disease-specific prediction in inherited cardiac conditions, and Itan et al. (2013) explored the human gene connectome for allele discovery, furthering the development of targeted analyses. Tools for Predicting Variant Pathogenicity Research is being conducted to improve variant effect prediction, focusing on phenotype-specific models. Several tools and databases are utilized in this effort, including the Human Gene Mutation Database (HGMD®) and ClinVar. Studies like Chen et al. (2024) and Backman et al. (2021) leverage large genomic datasets – 76,156 genomes and 454,787 UK Biobank participants respectively – to build mutational constraint maps and improve variant analysis. This work aims to more accurately identify disease-causing genetic variations. Multiple computational approaches are being employed to predict disease-gene associations and variant pathogenicity. Tools like CDG and VIPPID focus on specific diseases, such as primary immunodeficiency, to refine predictions. Other methods, including PHENOstruct and DeepPheno, utilize ontologies – like the Human Phenotype Ontology – to predict phenotypes from genetic variations. The goal is to improve both gene-level and protein-level feature association with pathogenic variants. Recent advances also include disease-specific prediction tools like DVPred for hearing loss and tools leveraging network-based methods for disease gene prediction as shown by Ata et al. (2021). Research also focuses on predicting recessive inheritance for missense variants, and improving prediction of Human Phenotype Ontology terms. These efforts demonstrate a growing focus on integrating diverse data types and computational methods to enhance variant interpretation and disease understanding. Disease-Specific Variant Prediction Approaches Disease-specific variant pathogenicity prediction is an area of active research, with tools being developed to improve variant interpretation in inherited conditions. Zhang et al. demonstrated significant improvements in prediction accuracy when using disease-specific approaches in inherited cardiac conditions. Further evidence of this trend is seen in tools like VIPPID, designed for primary immunodeficiency diseases, and DVPred, focused on hearing loss, indicating a move towards tailored prediction models. Several studies highlight the potential of network-based methods for disease gene prediction, as described by Ata et al. These methods, alongside tools utilizing human gene connectomes, aim to identify morbid alleles efficiently. The source also notes that prediction can be improved by incorporating features associated with gain-of-function and loss-of-function variants, as investigated by Sevim Bayrak et al., offering deeper insights into the functional impact of genetic changes. Researchers are also leveraging phenotype information to enhance prediction accuracy. Approaches like PHENOstruct, HPO2GO, and HPOLabeler utilize the Human Phenotype Ontology (HPO) to predict phenotype associations and improve variant prioritization. These methods, alongside tools like DeepPheno and HPOAnnotator, demonstrate the growing importance of integrating phenotypic data into computational models for accurate disease gene prediction. Identifying a High Fraction of the Human Genome to be under Selective Constraint Using GERP++.

Human Phenotype Ontology and Prediction Tools Research is being conducted to improve the prediction of variant effects, specifically utilizing phenotype-specific models. Several tools aim to predict Human Phenotype Ontology (HPO) terms, including PHENOstruct, HPO2GO, HPOLabeler, DeepPheno, HPOAnnotator, and others employing hierarchical or network-based approaches. These methods leverage diverse data sources and ontology information to associate genes and proteins with relevant phenotypic characteristics, enhancing disease gene prediction and understanding. Multiple computational approaches are being developed for disease-gene prediction, categorized by rationale and successes. These include tools focusing on network-based methods, disease-specific variant pathogenicity prediction (like VIPPID and DVPred for primary immunodeficiency or hearing loss), and methods leveraging HPO terms for improved annotation. The goal is to refine variant interpretation and prioritize potentially pathogenic alleles linked to specific diseases. Researchers are also exploring the use of the human gene connectome to discover morbid alleles. Approaches involve utilizing genomic mutational constraint maps derived from large datasets like 76,156 human genomes and exome sequencing of 454,787 UK Biobank participants. This work, along with the development of tools predicting HPO terms, contributes to a more comprehensive understanding of genotype-phenotype relationships and improved disease diagnosis. Source: https://www.nature.com/articles/s41467-025-66607-w_reference.pdf Tags: