Abstract
Artificial intelligence methods offer exciting new capabilities for the discovery of biological mechanisms from raw data because they are able to detect vastly more complex patterns of association that cannot be captured by classical statistical tests. Among these methods, deep neural networks are currently among the most advanced approaches and, in particular, convolutional neural networks (CNNs) have been shown to perform excellently for a variety of difficult tasks. Despite that applications of this type of networks to high-dimensional omics data and, most importantly, meaningful interpretation of the results returned from such models in a biomedical context remains an open problem. Here we present, an approach applying a CNN to nonimage data for feature selection. Our pipeline, DeepFeature, can both successfully transform omics data into a form that is optimal for fitting a CNN model and can also return sets of the most important genes used internally for computing predictions. Within the framework, the Snowfall compression algorithm is introduced to enable more elements in the fixed pixel framework, and region accumulation and element decoder is developed to find elements or genes from the class activation maps. In comparative tests for cancer type prediction task, DeepFeature simultaneously achieved superior predictive performance and better ability to discover key pathways and biological processes meaningful for this context. Capabilities offered by the proposed framework can enable the effective use of powerful deep learning methods to facilitate the discovery of causal mechanisms in high-dimensional biomedical data.
Highlights
Gene set selection facilitates interpretation of high-dimensional omics data by reducing it to components of greatest relevance and allowing them to be linked to specific functional themes and to extract meaningful clues for clarifying biological mechanisms
As illustrated by the analysis reported in this paper, DeepFeature-selected gene sets are both very different from more traditional approaches like lasso and analysis of variance (ANOVA), and appeared to be better aligned with meaningful biological mechanisms and consistently achieved higher enrichment for key pathways and functional groups
The results are produced by following an overall procedure of DeepFeature (Figure 1)
Summary
Gene set selection facilitates interpretation of high-dimensional omics data by reducing it to components of greatest relevance and allowing them to be linked to specific functional themes (e.g. using methods like process and pathway enrichment analysis) and to extract meaningful clues for clarifying biological mechanisms. Traditional machine learning (ML) algorithms (such as support vector machines [1], random forest, RF [2] and logistic regression [3]) are the ones most commonly applied in classification and feature selection (or gene selection) of nonimage data. Ever increasing data complexity is pushing the limits of ML algorithms to extract relevant information for phenotype identification related to disease diagnosis and analysis. In this respect, the selection of a small subset of critical elements or genes from a larger set has become a critical step. The element or gene selection ( known as feature selection) problem is not limited to genomic data analysis, but is an important process in many areas of research. The reliability of ML algorithms to find a subset of genes is mostly determined by the feature selection, feature extraction and classification steps
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
