Learning a hierarchical representation of the yeast transcriptomic machinery using an autoencoder model

Abstract

BackgroundA living cell has a complex, hierarchically organized signaling system that encodes and assimilates diverse environmental and intracellular signals, and it further transmits signals that control cellular responses, including a tightly controlled transcriptional program. An important and yet challenging task in systems biology is to reconstruct cellular signaling system in a data-driven manner. In this study, we investigate the utility of deep hierarchical neural networks in learning and representing the hierarchical organization of yeast transcriptomic machinery.ResultsWe have designed a sparse autoencoder model consisting of a layer of observed variables and four layers of hidden variables. We applied the model to over a thousand of yeast microarrays to learn the encoding system of yeast transcriptomic machinery. After model selection, we evaluated whether the trained models captured biologically sensible information. We show that the latent variables in the first hidden layer correctly captured the signals of yeast transcription factors (TFs), obtaining a close to one-to-one mapping between latent variables and TFs. We further show that genes regulated by latent variables at higher hidden layers are often involved in a common biological process, and the hierarchical relationships between latent variables conform to existing knowledge. Finally, we show that information captured by the latent variables provide more abstract and concise representations of each microarray, enabling the identification of better separated clusters in comparison to gene-based representation.ConclusionsContemporary deep hierarchical latent variable models, such as the autoencoder, can be used to partially recover the organization of transcriptomic machinery.

Extracting a biologically relevant latent space from cancer transcriptomes with variational autoencoders

“…7,8,32 For most features, the distribution of gene weights was similar: Many genes had weights near zero and few genes had high weights at each tail. In order to characterize patterns explained by selected encoded features of interest, we performed overrepresentation pathway analyses (ORA) separately for both positive and negative high weight genes; defined by greater than 2.5 standard deviations above or below the mean, respectively.…”

“…7,8,32 For instance, there were only 17 genes needed to identify patient sex ( Figure 3C). These genes were mostly located on sex chromosomes.…”

“…Reconstructing gene expression input data using autoencoder frameworks has been previously shown to reveal novel biological patterns. [7][8][9] VAEs and GANs are generative models, which means they learn to approximate a data generating distribution. Through approximation and compression, the models have been shown to capture an underlying data manifold -a constrained, lower dimensional space where data is distributed -and disentangle sources of variation from different classes of data.…”

Phenotypes Prediction from Gene Expression Data with Deep Multilayer Perceptron and Unsupervised Pre-training

“…It has been shown that the expression profiles regenerated by autoencoders improve the performance of gene clustering. In another study, unsupervised model like restricted Boltzmann machines and sparse autoencoders are used on yeast transcriptomic data [8]. By integrating knowledge from gene ontology and KEGG in the networks, they show that these models are capable of learning biologically sensible representations of the data and revealing novel insights regarding the machinery regulating gene expression.…”

scVAE: Variational auto-encoders for single-cell gene expression data

“…The latent representation is tuned to have low-capacity, so the model is forced to only estimate the most important features of the data. Different auto-encoder models have previously been used to model normalised (or binarised) bulk gene expression levels: denoising auto-encoders (Tan et al, 2014;Gupta et al, 2015;Tan et al, 2016), sparse auto-encoders (Chen et al, 2016), and robust autoencoders (Cui et al, 2017). A bottleneck auto-encoder (Eraslan et al, 2018) was also recently used to model single-cell transcript counts, and a generative adversarial network (Goodfellow et al, 2014), which is a related model, was recently used to model normalised single-cell gene expression levels (Ghahramani et al, 2018).…”