DeepIsoFun: a deep domain adaptation approach to predict isoform functions

Abstract

Motivation Isoforms are mRNAs produced from the same gene locus by alternative splicing and may have different functions. Although gene functions have been studied extensively, little is known about the specific functions of isoforms. Recently, some computational approaches based on multiple instance learning have been proposed to predict isoform functions from annotated gene functions and expression data, but their performance is far from being desirable primarily due to the lack of labeled training data. To improve the performance on this problem, we propose a novel deep learning method, DeepIsoFun, that combines multiple instance learning with domain adaptation. The latter technique helps to transfer the knowledge of gene functions to the prediction of isoform functions and provides additional labeled training data. Our model is trained on a deep neural network architecture so that it can adapt to different expression distributions associated with different gene ontology terms. Results We evaluated the performance of DeepIsoFun on three expression datasets of human and mouse collected from SRA studies at different times. On each dataset, DeepIsoFun performed significantly better than the existing methods. In terms of area under the receiver operating characteristics curve, our method acquired at least 26% improvement and in terms of area under the precision-recall curve, it acquired at least 10% improvement over the state-of-the-art methods. In addition, we also study the divergence of the functions predicted by our method for isoforms from the same gene and the overall correlation between expression similarity and the similarity of predicted functions. Availability and implementation https://github.com/dls03/DeepIsoFun/ Supplementary information Supplementary data are available at Bioinformatics online.

Computational Methods for Predicting Functions at the mRNA Isoform Level

“…Additionally, DeepIsoFun was compared to the other predictors on two more datasets- Arabidopsis thaliana and Drosophila melanogaster . DeepIsoFun outperformed the other predictors on these two datasets as well (Tables S2 and S3 from [ 31 ]). These comparisons were made using a small subset of 117 GO Slim terms which have been used in previous studies [ 51 ].…”

“…Previous methods have tried to address the issue of mRNA isoform function prediction by using a semi-supervised learning technique called MIL [ 30 , 36 , 37 , 38 ] However, the lack of labelled training data is reflected in their poor performance. To improve performance, DeepIsoFun [ 31 ] combines MIL with domain adaptation (DA) [ 50 ] to predict the functions of mRNA isoforms, using GO and RNA-Seq expression data.…”

“…WLRM [ 30 ]. In all the three datasets, DeepIsoFun outperformed the other predictors (Dataset #1 AUROC of 0.742 vs 0.64 (iMILP); Dataset #1 AUROC of 0.735 vs 0.679 (mi-SVM) and 0.69 (WLRM); Table 1 and Table 2 from [ 31 ]). Additionally, DeepIsoFun was compared to the other predictors on two more datasets- Arabidopsis thaliana and Drosophila melanogaster .…”

“…Previously, machine learning methods have been used to address a multitude of problems, some of which include, drug target discovery, gene function prediction, protein–protein interaction (PPI) prediction, protein structure and functional site prediction, and subcellular localization protein prediction [ 21 , 22 , 23 , 24 , 25 , 26 , 27 ]. More recently, several machine learning and recommendation system methods have also been developed to predict the biological functions of mRNA isoforms [ 28 , 29 , 30 , 31 , 32 , 33 , 34 , 35 , 36 , 37 , 38 ]. These methods have been successful in predicting gene functions at the level of mRNA isoforms and provide an added advantage over experimental approaches in terms of time and resources.…”

Isoform-Disease Association Prediction by Data Fusion

“…In fact, such lack also exists in functional analysis of isoforms [13]. To bypass this issue, some researchers take a gene as a bag and its spliced isoforms as instances of that bag, and adapt multiple instance learning (MIL) [2,17] to distribute the readily available functional annotations of a gene to its isoforms [3,6,14,26,34,40].…”