Describing mutational signatures using variational autoencoders
Authors
Term
4. term
Education
Publication year
2024
Submitted on
2024-06-10
Pages
21
Abstract
This report investigates the use of variational autoencoders for identifying mutational signatures within cancer genomics data. Mutational signatures represent characteristic patterns of mutations that can indicate underlying mutational processes, such as exposure to environmental factors or defects in DNA repair mechanisms. Traditional methods for extracting these signatures often employ Non-Negative Matrix Factorization (NMF). However, recent research explores the potential of autoencoders as a viable option within this field. This paper developed a β-VAE to find exposures, mutational signatures, and confidence intervals. The contribution of confidence intervals is unique to this paper and is derived by analyzing the probabilistic latent space. While experimental results demonstrate that theβ-VAE can achieve competitive performance, it lags behind state-of-the-art methods in terms of signature extraction. The findings highlight the need for further refinement and suggest future directions, including the exploration of alternative probabilistic models to enhance prediction accuracy.
Keywords
Documents