Deep learning-based cancer survival prognosis from RNA-seq data: approaches and evaluations

Abstract

Background: Recent advances in kernel-based Deep Learning models have introduced a new era in medical research. Originally designed for pattern recognition and image processing, Deep Learning models are now applied to survival prognosis of cancer patients. Specifically, Deep Learning versions of the Cox proportional hazards models are trained with transcriptomic data to predict survival outcomes in cancer patients. Methods: In this study, a broad analysis was performed on TCGA cancers using a variety of Deep Learning-based models, including Cox-nnet, DeepSurv, and a method proposed by our group named AECOX (AutoEncoder with Cox regression network). Concordance index and p-value of the log-rank test are used to evaluate the model performances. Results: All models show competitive results across 12 cancer types. The last hidden layers of the Deep Learning approaches are lower dimensional representations of the input data that can be used for feature reduction and visualization. Furthermore, the prognosis performances reveal a negative correlation between model accuracy, overall survival time statistics, and tumor mutation burden (TMB), suggesting an association among overall survival time, TMB, and prognosis prediction accuracy. Conclusions: Deep Learning based algorithms demonstrate superior performances than traditional machine learning based models. The cancer prognosis results measured in concordance index are indistinguishable across models while are highly variable across cancers. These findings shedding some light into the relationships between patient characteristics and survival learnability on a pan-cancer level.

Aberrant Chloride Intracellular Channel 4 Expression Is Associated With Adverse Outcome in Cytogenetically Normal Acute Myeloid Leukemia

“…Multi-omics analysis of the genetic and epigenetic mechanisms was important in leukemogenesis, including abnormal gene expression, microRNA-target gene interaction network analysis, and downstream gene ontology analysis (18,36). Our study further analyzed how CLIC4 overexpression affected the prognosis in CN-AML from these aspects.…”

TLSurv

“…Unlike traditional shallow learners, deep neural networks are able to capture sophisticated, non-linear data via activation functions and multiple levels of abstraction [20] [36]. Previous studies have shown that deep learning-based survival models have superior performance compared to shallow learners such as Cox-PH and RSF [4] [11] [15].…”

“…Since 2018, many deep learning methods; Cox-nnet [4], DeepSurv [15], AECOX [11] and Nnet-survival [7]; have been proposed. The first three adopt the CoxPH assumption by using a Cox regression layer as its output layer, while the last one supports non-proportional hazards.…”

“…The first three adopt the CoxPH assumption by using a Cox regression layer as its output layer, while the last one supports non-proportional hazards. Based on previous studies [7] [11], Cox-nnet has achieved the best performance among the four and it is suitable for small datasets. The Cox-nnet architecture is used in the embedding and survival section in this study.…”

The International Conference on Intelligent Biology and Medicine 2019 (ICIBM 2019): computational methods and applications in medical genomics

“…Huang et al developed a novel auto-encoder based model, namely AECOX, to identify prognostic marker genes from cohort transcriptomics data [14]. Comparing to classic models, AECOX utilized a novel auto-encoderbased formulation to derive non-linear features from the transcriptomics data that can well explain the low rank structure encoded in the data.…”