海角直播

>

VUStruct aims to make personalized structural biology faster with its automated pipeline for variants of unknown significance (VUS)

By Shelby Harris

The use of computational tools to provide a clinical diagnosis of genetic variants of unknown significance (VUSs) of rare diseases, in which the diagnosis is mainly informed by gene-phenotype association databases and DNA scoring methods, has been limited. Tools such as GERP, SIFT, and Polyphen have been shown to differentiate between known pathogenic/benign variants in large variant data sets, however, these methods process low specificity, leaving their abilities less useful for small sets of pre-filtered variants such as clinical case data.

Despite recent development of methods with increased predictive accuracy of scoring algorithms, there is still a looming risk of high false positive rates and the lack of 3D structural context of the affected protein. Furthermore, workflows that do take the 3D protein structure into context require a significant amount of human input as structures must be manually forwarded to various external web servers for specific calculations then explored with visualization tools.

In an effort to build upon these successes and tackle the challenges presented in earlier studies, Vanderbilt researchers Chris Moth and Jens Meiler created , denoted as VUStruct, to automate the various computational analyses that are needed to understand how a VUS may affect a structure of a protein. The researchers state that the goal of this pipeline is not to necessarily predict the pathogenicity of variants but more so as a method to provide more context and guidance for a user’s VUS. Following the user’s upload of their variant dataset to the VUStruct webserver, the jobs will be launched on Vanderbilt鈥檚 ACCRE computer cluster. VUStruct then integrates multiple 3D structures and transcripts, a novel method, and performs fresh calculations based on current databases.

The calculations include how the VUS may affect the free energy of protein folding (Rosetta 螖螖Gfolding), how it trends with known pathogenic/benign sites (PathProx), estimates the likelihood of the VUS affecting protein-protein interactions (ScanNet, PeSTo is also stated as a possible new method), predicts both protein post-translational modification (MusiteDeep) and digenic disease interactions (DigePred and DIEP). Finally, the pipeline provides the user with a case-wide report with the calculated results of their VUS dataset along with queried scores from Alphamissense, ConSurf, and COSMIS for each individual variant. Since the final audience of VUStruct includes clinicians, geneticists, and others who are not typically trained to interpret structural data and are more interested in simply knowing whether a VUS is worth pursuing, VUStruct鈥檚 final report includes a case summary that indicates the likelihood of a VUS being of interest. More detail on the 鈥渂ehind the scenes” computational work that VUStruct undergoes is available in the paper. It also includes more communication geared towards helping users like clinicians understand the process and its use in clinical case studies in the supporting information.

The researchers demonstrated VUStruct’s utility for hypothesis generation and structural analysis with two case studies: 1. It identified structural destabilization in the CACNA1D gene (F767L), leading to the hypothesis of a conformational shift toward an open-channel state to explain PASNA syndrome. 2. It uncovered a critical digenic interaction between the RPS19 and RPL27 genes, inspiring follow-up RNA sequencing that successfully diagnosed a complex case of Diamond-Blackfan anemia. Overall, across a subset of solved, single-VUS Vanderbilt Undiagnosed Diseases Network (UDN) cases, VUStruct consistently yielded actionable structural insights, including predictions for structural destabilization (74%), spatial clustering with known pathogenic variants (21%), and involvement in protein-protein interactions (13%).

VUStruct is available for download within the paper. While the researchers acknowledge the inability for VUStruct to generate multiple protein conformations, they hope to better the pipeline to enhance the structure selection algorithms for certain conformations, such as apo and holo. New models such as Boltz-2 and AlphaFold 3, should it become publicly available, can be included in the future pipeline versions.

Finally, they end with the note that the model’s prediction of digenic interactions could benefit from model retraining and that incorporation of recent availability novel datasets could improve VUStruct’s ability to predict disease caused by combined variants found in different genes.

 

Explore Story Topics

Recent News