Pancancer survival prediction using a deep learning architecture with multimodal representation and integration

Abstract

Motivation Use of multi-omics data carrying comprehensive signals about the disease is strongly desirable for understanding and predicting disease progression, cancer particularly as a serious disease with a high mortality rate. However, recent methods currently fail to effectively utilize the multi-omics data for cancer survival prediction and thus significantly limiting the accuracy of survival prediction using omics data. Results In this work, we constructed a deep learning model with multimodal representation and integration to predict the survival of patients using multi-omics data. We first developed an unsupervised learning part to extract high-level feature representations from omics data of different modalities. Then, we employed an attention-based method to integrate feature representations, produced by the unsupervised learning part, into a single compact vector and finally we fed the vector into fully connected layers for survival prediction. We used multimodal data to train the model and predict pancancer survival, and the results show that using multimodal data can lead to higher prediction accuracy compared to using single modal data. Furthermore, we used the concordance index and the 5-fold cross-validation method for comparing our proposed method with current state-of-the-art methods and our results show that our model achieves better performance on the majority of cancer types in our testing datasets. Availability https://github.com/ZhangqiJiang07/MultimodalSurvivalPrediction

Survival Prediction Landscape: An In-Depth Systematic Literature Review on Activities, Methods, Tools, Diseases, and Databases

“…Table 2 illustrates disease specific predictors distribution for both cancer and other diseases respectively. In the last 3 years, 60 predictors have been designed for different cancer subtypes related survival prediction 24,104,108 while only 14 predictors have been designed for other diseases such as cardiovascular diseases, COVID-19, and trauma 29,112,119,120 .…”

“…The choice of omics type hinges on the specific disease under investigation, as indicated by the disease-wise distribution of omics types in Figure 5. Out of 9 omics types, mRNA, CNV, miRNA, and methylation have been the most commonly utilized modalities for 33 cancer subtypes i.e., breast cancer 14,23,68,74,90,[98][99][100] , pancancer 24,91,[105][106][107][108]131 , colon can- cer 39,[74][75][76][77] , lung adenocarcinoma 27,101,102 , and ovarian cancer 72,84,[88][89][90]103 . In addition, mutation data has been utilized for 7 cancer subtypes namely, adult diffuse glioma 69 , breast cancer 23 , cervical cancer 73 , non-small cell lung cancer 95 , ovarian cancer 103 , and pancreatic cancer 32 .…”

“…To expedite advancements in survival prediction research, researchers are harnessing the capabilities of AI algorithms by utilizing extensive survival-related data from public databases such as the Cancer Genome Atlas Program (TCGA) 17 , and NCI Genomic Data Commons (GDC) 1819–24 . In addition, the diversity and heterogeneity of diseases hinder the development of a universally applicable survival prediction pipeline 14,25 .…”

Dual-space disentangled-multimodal network (DDM-net) for glioma diagnosis and prognosis with incomplete pathology and genomic data

“…In the fusion stage, they compressed different modalities into a single feature vector for each patient and scaled up the weights of available modalities for incomplete cases. Fan et al (2023) improved upon this model by revising the unsupervised loss and incorporating an attention mechanism to automatically assign weights to different modalities. Vale Silva and Rohr (2020) also applied the attention mechanism and proposed a model named MultiSurv for risk prediction for 33 cancer types.…”

“…The statistic information of the TCGA-GBM/LGG project. Comparison with the state-of-the-art methods In this section, we present the results of our proposed DDM-net framework and five state-of-the-art methods including CPM (Zhang et al 2019a, Cui et al 2022b), multimodal prognosis (MP)(Cheerla and Gevaert 2019), multimodal survival prediction (MSP)(Fan et al 2023), MMD (Cui et al 2022a) and disentangled-multimodal adversarial autoencoder (DMM-AAE).…”