2023 ACMG Annual Clinical Genetics Meeting Digital Edition: Poster Gallery-Getting it Right on the First Test: Machine Learning Plus Genome-wide Methylation Profiling Resolves Equivocal Cases of Beckwith-Wiedemann Syndrome

Pdf Summary

Researchers at the Mayo Clinic have developed a machine learning pipeline coupled with a genome-wide DNA methylation array to diagnose cases of Beckwith-Wiedemann Syndrome (BWS) that were previously equivocal by methylation-specific multiplex-ligation dependent probe amplification (MS-MLPA) in a single attempt. BWS is a growth disorder caused by epigenetic alterations on chromosome 11p15.5. The new approach utilizes whole genome methylation analysis to detect all epigenomic abnormalities and applies adjusted reference ranges for methylation quantification. The study cohort included DNA samples from a control group of normal patients, a training group of patients with known methylation disorders (including BWS cases), and an equivocal group of cases from the MS-MLPA assay. The samples were run on the Illumina Infinium MethylationEPIC V1 850K Array and analyzed using the R ChAMP library. The data underwent quality control, normalization, and outlier analysis using a generalized additive model. Dimensionality reduction was then performed using the UMAP algorithm and target regions. The researchers employed Autogluon for cross-validation and hyperparameter tuning of classification algorithms. An ensemble model was used to provide a final disease classification. Additionally, residual clinical samples from 10 equivocal cases were analyzed using the machine learning classifier to determine if it could resolve borderline calls. The results showed that 70% of the equivocal samples were correctly diagnosed as BWS by the classifier, while the remaining cases could not be resolved due to limited positive samples, inaccurate MS-MLPA results, or somatic mosaicism. The study demonstrates the potential of machine learning and whole genome methylation analysis to improve the sensitivity and specificity of BWS testing.

Asset Subtitle

Presenting Author - Jeanne Theis, PhD; Co-Author - Jayson J. Hardcastle, Ph.D.; Co-Author - Jason Vollenweider, BS; Co-Author - Calvin Jerde, MS; Co-Author - Kandelaria M. Rumilla, MD; Co-Author - Christine M. Koellner, M.S.; Co-Author - Eric Klee; Co-Author - Jesse R. Walsh, PhD; Co-Author - Garrett Jenkinson, PhD; Co-Author - Jagadheshwar Balan, MS; Co-Author - Linda Hasadsri, MD, PhD, FACMG;

Meta Tag

Bioinformatics

Epigenetics

Genomic Methodologies

Imprinting

Methodology

Methylation

Microarray

Co-Author Jayson J. Hardcastle, Ph.D.

Co-Author Jason Vollenweider, BS

Co-Author Calvin Jerde, MS

Co-Author Kandelaria M. Rumilla, MD

Co-Author Christine M. Koellner, M.S.

Co-Author Eric Klee

Co-Author Jesse R. Walsh, PhD

Co-Author Garrett Jenkinson, PhD

Co-Author Jagadheshwar Balan, MS

Co-Author Linda Hasadsri, MD, PhD, FACMG

Presenting Author Jeanne Theis, PhD

Keywords

Mayo Clinic

machine learning

genome-wide DNA methylation array

Beckwith-Wiedemann Syndrome

methylation-specific multiplex-ligation dependent probe amplification

epigenetic alterations

chromosome 11p15.5

whole genome methylation analysis

Illumina Infinium MethylationEPIC V1 850K Array

R ChAMP library