AutoGenome V2: New Multimodal Approach Developed for Multi-Omics Research

Abstract

Deep learning is very promising in solving problems in omics research, such as genomics, epigenomics, proteomics, and metabolics. The design of neural network architecture is very important in modeling omics data against different scientific problems. Residual fully-connected neural network (RFCN) was proposed to provide better neural network architectures for modeling omics data. The next challenge for omics research is how to integrate informations from different omics data using deep learning, so that information from different molecular system levels could be combined to predict the target. In this paper, we present a novel multimodal approach that could efficiently integrate information from different omics data and achieve better accuracy than previous approaches. We evaluate our method in four different tasks: drug repositioning, target gene prediction, breast cancer subtyping and cancer type prediction, and all the four tasks achieved state of art performances.The multimodal approach is implemented in a software package called AutoOmics to facilitate researchers to use.

AutoGGN: A Gene Graph Network AutoML Tool for Multi-Omics Research

“…The classification accuracy was 0.982, which was higher than the accuracy 0.968 achieved using single-omics and interaction network. Therefore, AutoGGN outperformed AutoOmics 10 by one percentage point (Figure 2A), which used multi-omics data as input solely. The detail accuracy for each cancer type was showed in a confusion matrix of AutoGGN (Figure 2B).…”

“…Fully-connected neural network showed powerful ability in analyzing omics data and achieved better performance in some tasks 9,10 , but its ability of handle biological networks is limited. AutoGGN, a multimodal method using graph convolutional neural network proposed by us, could integrate molecular interaction networks with omics data (Figure 1) efficiently.…”

“…It could also be implemented to solve various biological and medical problems (Figure 1B), including patient stratification, patient survival prediction, biomarker identification and drug sensitivity prediction. In this study, we evaluated AutoGGN on two classification tasks in biology and compared the model performance with other published methods, such as MLP in AutoKeras 19 , RFCN in AutoGenome 9 , AutoOmics 10 . The first case is a cancer type classification task, which involves 9,000 patient samples covering 24 cancer types and two types of omics data (gene mutation and gene expression).…”

“…When using gene expression data and interaction network data as input for the model, AutoGGN achieved an accuracy of 0.968, which was much higher than XGBoost 21 and AutoKeras 19 . When using multi-omics profiles (gene mutation and gene expression) and network data as input, AutoGGN also reached a high accuracy of 0.982 and outperformed AutoOmics 10 by one percentage point, which was designed to take multi-omics data as input. In single-cell stage classification task, our method demonstrated highest performance among other published models, proving AutoGGN’s capability to make full use of the information from omics features and their interactions during training.…”

“…Deep learning (DL) has demonstrated potential in mining complex and heterogeneous biological data 7 . DL networks such as feedforward fully-connected neural network (FFNN) and randomly-wired residual fully-connected neural network (RRFCN) have already been proven as efficient approaches on omics data analysis 8–10 . These algorithms are powerful for interpreting omics data through fully-connected neural networks, which can form a classification decision from samples with omics features 9,10 .…”