Predicting clinical outcomes from large scale cancer genomic profiles with deep survival models

Abstract

Translating the vast data generated by genomic platforms into accurate predictions of clinical outcomes is a fundamental challenge in genomic medicine. Many prediction methods face limitations in learning from the high-dimensional profiles generated by these platforms, and rely on experts to hand-select a small number of features for training prediction models. In this paper, we demonstrate how deep learning and Bayesian optimization methods that have been remarkably successful in general high-dimensional prediction tasks can be adapted to the problem of predicting cancer outcomes. We perform an extensive comparison of Bayesian optimized deep survival models and other state of the art machine learning methods for survival analysis, and describe a framework for interpreting deep survival models using a risk backpropagation technique. Finally, we illustrate that deep survival models can successfully transfer information across diseases to improve prognostic accuracy. We provide an open-source software implementation of this framework called SurvivalNet that enables automatic training, evaluation and interpretation of deep survival models.

A convolutional neural network to filter artifacts in spectroscopic MRI

“…The CNNs can be treated as a complex mathematical function with some unknown underlying statistical model. Bayesian optimization assumes that a series of Gaussian distributions can reproduce the underlying model and uses those to indirectly optimize the CNN; this technique has previously been successfully applied to tune architecture hyperparameters . A Bayesian optimization approach using the Spearmint framework was used in this work to tune the size of the convolution kernels and the dropout fraction of the fully connected layers .…”

“…Thus, these features capture criteria that MRS experts explicitly believe are important to spectral quality, and machine‐learning algorithms built using these engineered features have shown promise as spectral quality filters . In contrast to systems that model expert beliefs explicitly, deep learning broadly defines a category of machine learning algorithms that can learn underlying features from raw data without the need for any a priori definition of such features, and have been able to shatter benchmarks in natural language processing, medical image segmentation, survival analysis, and identification of pathology in medical images . Deep convolutional neural networks (CNNs) in particular are well suited to analyzing waveforms similar to raw spectra, and have only recently been applied in the context of MRS .…”

Extreme learning machine Cox model for high‐dimensional survival analysis

“…(a) Choose an appropriate kernel function and calculate the kernel matrix K n × n as in (9). (b) Replace the linear function in (11) by K n × n and similar to the kernel Cox regression, 37 we obtain the following partial log likelihood for ELMCoxBAR model:…”

“…7–9,22–24 But perhaps due to the computation burden incurred by the backpropagation algorithms that are widely employed in both multilayer perceptron (MLP) neural networks and deep neural networks, most of these proposals focus on low-dimensional survival data. These sophisticated deep learning methods have demonstrated their usefulness in high-dimensional setting.…”

“…[1][2][3][4][5][6] However, the model assumptions underlying these models, such as the proportional hazard, independence between censoring and survival times, linearity in coefficients, may not be tenable in real applications. For this reason, nonparametric approaches such as neural networks [7][8][9] can be useful alternatives. In particular, neural networks methodology has been adapted to survival analysis for more than two decades.…”