A deep embedded refined clustering approach for breast cancer distinction based on DNA methylation

Abstract

Epigenetic alterations have an important role in the development of several types of cancer. Epigenetic studies generate a large amount of data, which makes it essential to develop novel models capable of dealing with large-scale data. In this work, we propose a deep embedded refined clustering method for breast cancer differentiation based on DNA methylation. In concrete, the deep learning system presented here uses the levels of CpG island methylation between 0 and 1. The proposed approach is composed of two main stages. The first stage consists in the dimensionality reduction of the methylation data based on an autoencoder. The second stage is a clustering algorithm based on the soft assignment of the latent space provided by the autoencoder. The whole method is optimized through a weighted loss function composed of two terms: reconstruction and classification terms. To the best of the authors’ knowledge, no previous studies have focused on the dimensionality reduction algorithms linked to classification trained end-to-end for DNA methylation analysis. The proposed method achieves an unsupervised clustering accuracy of 0.9927 and an error rate (%) of 0.73 on 137 breast tissue samples. After a second test of the deep-learning-based method using a different methylation database, an accuracy of 0.9343 and an error rate (%) of 6.57 on 45 breast tissue samples are obtained. Based on these results, the proposed algorithm outperforms other state-of-the-art methods evaluated under the same conditions for breast cancer classification based on DNA methylation data.

Deep Learning for Human Disease Detection, Subtype Classification, and Treatment Response Prediction Using Epigenomic Data

“…Applications of DL to assist physicians and scientists in clinical settings using epigenomic data have been relatively unexplored until very recently. Except for one paper published in 2016, 21 out of 22 papers reviewed were published in the last 5 years and a majority of the models were developed by USA and China research teams [ 21 , 22 , 23 , 24 , 25 , 26 , 27 , 28 , 29 , 30 , 31 , 32 , 33 , 34 , 35 , 36 , 37 , 38 , 39 , 40 , 41 , 42 ]. This suggests a novel field of study that gains an increasing interest.…”

“…Amor et al [ 25 ] and Si et al [ 29 ] developed two-stage models to identify cancer samples among normal samples using DNA methylation data. In both models, the dimensionality reduction was performed using autoencoder (AE) structures which allowed to extract features automatically.…”

“…In both models, the dimensionality reduction was performed using autoencoder (AE) structures which allowed to extract features automatically. Si et al [ 29 ] grouped extracted the features into cancer or non-cancer using k-means, Gaussian mixture method, and self-organizing map (SOM), while del Amor et al [ 25 ] proposed a novel approach called deep embedded refined clustering (DERC) that trained end-to-end optimizing the dimensionality reduction and the unsupervised classification in the same step. These studies [ 25 , 29 ] found the followings: DNN-based extracted features were more effective for clustering analysis than those extracted from the principal component analysis (PCA) and non-negative matrix factorization (NMF), DNN and SOM outperformed previous probabilistic mixture methods [ 44 , 45 , 46 ], DERC achieved higher accuracy in breast cancer classification in comparison with other models under the same conditions.…”

“…DNA methylation has been the most investigated epigenetic mechanism because of its roles on gene expression regulation including X-chromosome inactivation and allele-specific silencing of imprinted genes that are preferentially expressed from only one of the parental copies [ 48 , 49 ]. Integration between traditional biochemical methodologies and novel bioinformatic analysis methods such as DL extended our understanding for DNA methylation patterns in various types of human diseases such as cancer [ 25 , 26 , 28 , 29 , 30 , 33 , 34 , 36 , 39 , 40 ], concussion [ 21 ], schizophrenia [ 28 ], and cardiovascular diseases [ 20 ]. Profiling DNA methylation at a genome-wide level could be conducted using various types of sequencing technologies.…”

“…There are three categories of commonly used supervised feature selection methods including filter, wrapper, and embedded [ 31 ]. For example, t -test, Wilcoxon rank sum, and F-test with a threshold of p -value were applied to filter significant biomarkers [ 25 , 27 , 30 , 42 ]. Subsequently, the p -value was adjusted using the Benjamini–Hochberg correction for false discovery rate [ 23 , 24 , 28 ].…”