Integrated multi-omics analysis of ovarian cancer using variational autoencoders

Abstract

Cancer is a complex disease that deregulates cellular functions at various molecular levels (e.g., DNA, RNA, and proteins). Integrated multi-omics analysis of data from these levels is necessary to understand the aberrant cellular functions accountable for cancer and its development. In recent years, Deep Learning (DL) approaches have become a useful tool in integrated multi-omics analysis of cancer data. However, high dimensional multi-omics data are generally imbalanced with too many molecular features and relatively few patient samples. This imbalance makes a DL based integrated multi-omics analysis difficult. DL-based dimensionality reduction technique, including variational autoencoder (VAE), is a potential solution to balance high dimensional multi-omics data. However, there are few VAE-based integrated multi-omics analyses, and they are limited to pancancer. In this work, we did an integrated multi-omics analysis of ovarian cancer using the compressed features learned through VAE and an improved version of VAE, namely Maximum Mean Discrepancy VAE (MMD-VAE). First, we designed and developed a DL architecture for VAE and MMD-VAE. Then we used the architecture for mono-omics, integrated di-omics and tri-omics data analysis of ovarian cancer through cancer samples identification, molecular subtypes clustering and classification, and survival analysis. The results show that MMD-VAE and VAE-based compressed features can respectively classify the transcriptional subtypes of the TCGA datasets with an accuracy in the range of 93.2-95.5% and 87.1-95.7%. Also, survival analysis results show that VAE and MMD-VAE based compressed representation of omics data can be used in cancer prognosis. Based on the results, we can conclude that (i) VAE and MMD-VAE outperform existing dimensionality reduction techniques, (ii) integrated multi-omics analyses perform better or similar compared to their mono-omics counterparts, and (iii) MMD-VAE performs better than VAE in most omics dataset.

Figure 1

Methods: (A) VAE/MMD-VAE architecture consists of an encoder and a decoder made from 3 hidden layers and a bottleneck made from 2 layers and a 3-layered ANN-based classifier for supervised LFs learning, (B) Clustering using 2 LFs and ANN-based classification (e.g., cancer vs normal, and molecular subtypes) using 2 and 128 LFs, (C) Survival analysis using 128 LFs: (i) inferring survival subgroup, (ii) predicting subgroup and (iii) potential prognostic biomarkers.

Figure 2

Clustering of normal and cancer samples using the LFs learned using unsupervised PCA, t-SNE, VAE & MMD-VAE (using 2D for PCA & t-SNE and first 2 LFs for VAE and MMD-VAE) (a)–(d)) on DNA methylation (mono-omics) data from the GDC cohort. t-SNE was used (e,f) on the 128 learned LFs to identify 2 LFs for the clustering. Legends: 0—Normal, 1—Cancer.

Figure 3

Clustering molecular subtypes using the LFs learned through the supervised VAE & MMD-VAE + t-SNE (2D or 2 LFs): (a–c) for MMD-VAE respectively for mono-omics, di-omics and tri-omics data, (d–f) for MMD-VAE + t-SNE respectively for mono-omics, di-omics and tri-omics data. Legends: 0—Immunoreactive, 1—Differentiated, 2—Proliferative and 3—Mesenchymal.

Figure 4

Survival analysis using existing using molecular subtypes and CRLFs-based survival subgroups: (a) survival analysis using the existing transcriptional subtypes show that they are not linked to the survival ($p=0.19<0.05$), (b–f) survival analysis using the two subgroups show significant survival differences ($p<0.05$) between the groups. The results in (e) for 292 samples, and the rest are for 481 samples.

Figure 5

Association between CRLFs and input features: Input features of CNV_mRNA_methylation omics data are clustered based on the correlation data with the identified CRLFs. For example, the NDRG2 gene has strong correlation with LF30 and LF69.