Semi-supervised learning of the electronic health record for phenotype stratification

Abstract

Patient interactions with health care providers result in entries to electronic health records (EHRs). EHRs were built for clinical and billing purposes but contain many data points about an individual. Mining these records provides opportunities to extract electronic phenotypes, which can be paired with genetic data to identify genes underlying common human diseases. This task remains challenging: high quality phenotyping is costly and requires physician review; many fields in the records are sparsely filled; and our definitions of diseases are continuing to improve over time. Here we develop and evaluate a semi-supervised learning method for EHR phenotype extraction using denoising autoencoders for phenotype stratification. By combining denoising autoencoders with random forests we find classification improvements across multiple simulation models and improved survival prediction in ALS clinical trial data. This is particularly evident in cases where only a small number of patients have high quality phenotypes, a common scenario in EHR-based research. Denoising autoencoders perform dimensionality reduction enabling visualization and clustering for the discovery of new subtypes of disease. This method represents a promising approach to clarify disease subtypes and improve genotype-phenotype association studies that leverage EHRs.

Unsupervised Extraction of Stable Expression Signatures from Public Compendia with an Ensemble of Neural Networks

“…To tap this potential, we will require algorithms like eADAGE that robustly integrate these diverse datasets in a manner that is not limited to well-understood aspects of biology. Furthermore, while public compendia tend to be dominated by expression data, autoencoders have also been successfully applied to datasets based on large collections of electronic health records where they are particularly effective at dealing with missing data (Beaulieu-Jones et al, 2016; Miotto et al, 2016, Beaulieu-Jones and Moore, 2017). These features, along with their unsupervised nature, make DAs a promising approach for the integration of heterogeneous data types.…”

DeepImpute: an accurate, fast and scalable deep neural network method to impute single-cell RNA-Seq data

“…As reflected by the name, it belongs to the class of deep neural-network models (Ching, Zhu, et al , 2018;Alakwaa et al , 2018;Chaudhary et al , 2018) . Recent years, deep learning and related deep neural network algorithms have gained much interest in the biomedical field (Ching, Himmelstein, et al , 2018) , ranging from applications from extracting stable gene expression signa tures in large sets of public data (Tan et al , 2017) to stratify phenotypes (Beaulieu-Jones et al , 2016) or impute missing values (Beaulieu-Jones and Moore, 2017) using electronic health record (EHR) data. Here, we construct DeepImpute models by splitting the genes into subsets and builds sub-networks to increase its efficacy and efficienc y.…”

Reproducibility of computational workflows is automated using continuous analysis

“…Interactive development tools, such as Jupyter 19,20 , RMarkdown 21,22 and Sweave 23 can be incorporated to present the code and analysis in a logical graphical manner. For example, we recently used Jupyter with continuous analysis in our own publication 24 and corresponding repository 25 . Reviewers can follow what was done in an audit fashion without having to install and run software while having confidence that analyses are reproducible.…”