Exploring Machine Learning Approaches for Phenotype Prediction of Huntington's Disease - Final Degree Project

Aug 5, 2026

Upload by Aurora

Huntington’s disease symptom onset is largely predicted using the CAG trinucleotide expansion length in the HTT gene, yet this explains only about half of the phenotypic variability, while a substantial portion is considered heritable. Genome-wide association studies have suggested genetic modifiers, though often limited by linear-effect assumptions and computational demands. This project benchmarks machine learning models on an EnrollHD GWAS dataset to predict age at disease onset, comparing regularized linear, tree-based approaches, and an ordinary least squares baseline, and extracts key predictive SNP features to identify potential modifiers.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages79

Reading Time3 h 19 min

AccessFree to download

Abstract

Acknowledgements

List of Figures

List of Tables

Final Degree Project Biomedical Engineernig Degree “Exploring machine learning approaches for phenotype prediction of Huntington's disease’’ Barcelona, 5th of May, 2024 Author: Caterina Fuses i Kuzmina Director: Jordi Abante Llenas Tutor: Josep Maria Canals Coll Abstract Huntington’s disease onset of symptoms is clinically predicted primarily using the length of the CAG trinucleotide expansion in the HTT gene. However, this prediction can only explain around 50% of the variability of the phenotype. It is estimated that 40% of the remaining variability is heritable, suggesting the presence of other genetic factors. Genome Wide Association Studies (GWAS) have identified potential genetic modifiers, although only through the revelation of linear effects and via computationally demanding processes. This project benchmarks various machine learning algorithms trained with an EnrollHD GWAS dataset to predict the age at HD onset. The dataset comprises the genotype of millions of SNPs from approximately 9,000 individuals. The models considered include regularized linear models (Lasso and Elastic Net) and tree-based models (Random Forest and XGBoost), and their predictive power is compared to an Ordinary Least Squares baseline model trained solely with sex and CAG as covariates. The results indicate that tree-based models achieve the best estimation of age of onset (AO), improving the prediction by 3% with respect to the baseline, possibly due to their implicit consideration of interactions between SNPs. For each model, we extract the most significant features contributing to the model, thereby identifying genetic modifiers. Some of these key SNPs are in well-known AO modifier candidates such as FAN1 and MYT1L, while others are in genes like CDYL2 proposed as new candidates. Keywords: Machine Learning, High Dimensional Data, Huntington’s Disease, Single Nucleotide Polymorphism. Acknowledgements The most important person to acknowledge in this project is my director, Jordi Abante. The project originated from his innovative idea for a novel research path to develop abetter AO prediction algorithm. I am deeply honored to have contributed to the early stages of this endeavor with this initial exploration. My utmost gratitude goes to him for his constant guidance and indispensable insights at every step. I have enjoyed working with him enormously and am very grateful to have completed this project under his supervision. List of Figures 1 Inverse correlation of age of onset and CAG repeat length observed in the Enroll-HD dataset used in this project................. 1 2 Cumulative probability of onset of HD for various CAG lengths based on the Langbehn model. Figure by Langbehn et al. [30] ......... 6 3 Mean as estimated by various published formulae. Figure by Langbehn, Hayden, Paulsen et al. [31] ......................... 7 4 Project’s general workflow.......................... 10 5 Data preprocessing flow chart........................ 19 6 Graphical normality tests.......................... 25 7 Comparison of least squares method prediction (Lasso), top image, and tree-based method prediction (Histogram XGBoost), bottom image... 33 8 Manhattan plots of a least squares method (Lasso), top image, and a tree-based method (Histogram XGBoost), bottom image......... 34 9 AO over CAG repeat length of 500 random samples, colored by the presence of an alternative allele in SNPs rs61997076 (left) and rs144287831 (right) ..................................... 35 10 AO over CAG repeat length of 500 random samples, colored by the presence of an alternative allele in SNPs rs118089305 (left) and rs10169129 (right) ..................................... 36 11 AO over CAG repeat length of 500 random samples, colored by the presence of an alternative allele in rs141338757 .............. 37 12 GWAS by Lee et al. [8], showing significant associations between AO and different chromosome loci........................

FAQs

Why does CAG repeat length only partially predict Huntington’s disease phenotype?

CAG repeat length in the HTT gene explains only around 50% of phenotypic variability. The remaining variability is estimated to be heritable, implying additional genetic contributors.

Which machine learning models were benchmarked for predicting age at onset (AO)?

The project compares regularized linear models (Lasso, Elastic Net), tree-based models (Random Forest, XGBoost), and an ordinary least squares baseline using sex and CAG as covariates.

What improvement did the best models achieve over the baseline, and how were genetic modifiers identified?

Tree-based models produced the best AO estimation, improving prediction by about 3% versus the baseline. For each model, the most significant features were extracted to identify likely genetic modifier SNPs.

Aug 5, 2026

Upload by Aurora