Imputing missing RNA-sequencing data from DNA methylation by using a transfer learning–based neural network

Abstract

<jats:title>Abstract</jats:title> <jats:sec> <jats:title>Background</jats:title> <jats:p>Gene expression plays a key intermediate role in linking molecular features at the DNA level and phenotype. However, owing to various limitations in experiments, the RNA-seq data are missing in many samples while there exist high-quality of DNA methylation data. Because DNA methylation is an important epigenetic modification to regulate gene expression, it can be used to predict RNA-seq data. For this purpose, many methods have been developed. A common limitation of these methods is that they mainly focus on a single cancer dataset and do not fully utilize information from large pan-cancer datasets.</jats:p> </jats:sec> <jats:sec> <jats:title>Results</jats:title> <jats:p>Here, we have developed a novel method to impute missing gene expression data from DNA methylation data through a transfer learning–based neural network, namely, TDimpute. In the method, the pan-cancer dataset from The Cancer Genome Atlas (TCGA) was utilized for training a general model, which was then fine-tuned on the specific cancer dataset. By testing on 16 cancer datasets, we found that our method significantly outperforms other state-of-the-art methods in imputation accuracy with a 7–11% improvement under different missing rates. The imputed gene expression was further proved to be useful for downstream analyses, including the identification of both methylation–driving and prognosis-related genes, clustering analysis, and survival analysis on the TCGA dataset. More importantly, our method was indicated to be useful for general purposes by an independent test on the Wilms tumor dataset from the Therapeutically Applicable Research to Generate Effective Treatments (TARGET) project.</jats:p> </jats:sec> <jats:sec> <jats:title>Conclusions</jats:title> <jats:p>TDimpute is an effective method for RNA-seq imputation with limited training samples.</jats:p> </jats:sec>

Integrative Analysis of Next-Generation Sequencing for Next-Generation Cancer Research toward Artificial Intelligence

“…For instance, TDimpute aims at generating missing RNA-seq data by training a DNN on methylation data. They used TCGA and TARGET (https://ocg.cancer.gov/programs/target/data-matrix, accessed on 20 May 2021) data as proof of concept of the applicability of DNN for data imputation in a multi-omics integration study [120]. Because this integrative model can exploit information in different levels of regulatory mechanisms, it can build a more detailed model and achieve better performance than a model build on a single-omics dataset [117,121].…”

An Adaptive Transfer-Learning-Based Deep Cox Neural Network for Hepatocellular Carcinoma Prognosis Prediction

“…Yet, these methods are limited due to cancer data from small sample sizes (11,12). One common strategy to solve this problem is transfer learning, where models are pre-trained based on similar cancer types and then fine-tuned on the target cancer type (13,14). For example, Vanacker used the messenger RNA (mRNA) expression data collected from The Cancer Genome Atlas (TCGA) to train the initial deep neural network and fine-tune it for patient risk estimation (15).…”

OmiTrans: generative adversarial networks based omics-to-omics translation framework

“…However, they came up with the conclusion that DNA methylation data had limited prediction power for gene expression, which is mainly because of the rather low capacity of their network according to our experiments. TDimpute [Zhou et al, 2020] is a more recent work in this field that applied a quite straightforward neural network to impute missing RNA-Seq data from DNA methylation data. They claimed that TDimpute significantly outperformed other stateof-the-art methods including singular value decomposition imputation (SVD), trans-omics block missing data imputation (TOBMI) [Dong et al, 2019], and LASSO [Tibshirani, 1996].…”

“…They claimed that TDimpute significantly outperformed other stateof-the-art methods including singular value decomposition imputation (SVD), trans-omics block missing data imputation (TOBMI) [Dong et al, 2019], and LASSO [Tibshirani, 1996]. TDimpute [Zhou et al, 2020] is currently the most state-ofthe-art method that is able to predicting gene expression from DNA methylation, and it is our major comparing method.…”