Interpretable deep generative models for genomics

Deep neural networks implementing generative models for dimensionality reduction have been extensively used for the visualization and analysis of genomic data. One of their key limitations is lack of interpretability: it is challenging to quantitatively identify which input features are used to construct the embedding dimensions, thus preventing insight into why cells are organized in a particular data visualization, for example. Here we present a scalable, interpretable variational autoencoder (siVAE) that is interpretable by design: it learns feature embeddings that guide the interpretation of the cell embeddings in a manner analogous to factor loadings of factor analysis. siVAE is as powerful and nearly as fast to train as the standard VAE but achieves full interpretability of the embedding dimensions. We exploit a number of connections between dimensionality reduction and gene network inference to identify gene neighborhoods and gene hubs, without the explicit need for gene network inference. Finally, we observe a systematic difference in the gene neighborhoods identified by dimensionality reduction methods and gene network inference algorithms in general, suggesting they provide complementary information about the underlying structure of the gene co-expression network.

Transcriptomic time-series analysis of cold- and heat-shock response in psychrotrophic lactic acid bacteria

Background Psychrotrophic lactic acid bacteria (LAB) species are the dominant species in the microbiota of cold-stored modified-atmosphere-packaged food products and are the main cause of food spoilage. Despite the importance of psychrotrophic LAB, their response to cold or heat has not been studied. Here, we studied the transcriptome-level cold- and heat-shock response of spoilage lactic acid bacteria with time-series RNA-seq for Le. gelidum, Lc. piscium, and P. oligofermentans at 0 °C, 4 °C, 14 °C, 25 °C, and 28 °C. Results We observed that the cold-shock protein A (cspA) gene was the main cold-shock protein gene in all three species. Our results indicated that DEAD-box RNA helicase genes (cshA, cshB) also play a critical role in cold-shock response in psychrotrophic LAB. In addition, several RNase genes were involved in cold-shock response in Lc. piscium and P. oligofermentans. Moreover, gene network inference analysis provided candidate genes involved in cold-shock response. Ribosomal proteins, tRNA modification, rRNA modification, and ABC and efflux MFS transporter genes clustered with cold-shock response genes in all three species, indicating that these genes could be part of the cold-shock response machinery. Heat-shock treatment caused upregulation of Clp protease and chaperone genes in all three species. We identified transcription binding site motifs for heat-shock response genes in Le. gelidum and Lc. piscium. Finally, we showed that food spoilage-related genes were upregulated at cold temperatures. Conclusions The results of this study provide new insights on the cold- and heat-shock response of psychrotrophic LAB. In addition, candidate genes involved in cold- and heat-shock response predicted using gene network inference analysis could be used as targets for future studies.

Improving gene network inference with graph wavelets and making insights about ageing-associated regulatory changes in lungs

Using gene-regulatory-networks-based approach for single-cell expression profiles can reveal unprecedented details about the effects of external and internal factors. However, noise and batch effect in sparse single-cell expression profiles can hamper correct estimation of dependencies among genes and regulatory changes. Here, we devise a conceptually different method using graphwavelet filters for improving gene network (GWNet)-based analysis of the transcriptome. Our approach improved the performance of several gene network-inference methods. Most Importantly, GWNet improved consistency in the prediction of gene regulatory network using single-cell transcriptome even in the presence of batch effect. The consistency of predicted gene network enabled reliable estimates of changes in the influence of genes not highlighted by differential-expression analysis. Applying GWNet on the single-cell transcriptome profile of lung cells, revealed biologically relevant changes in the influence of pathways and master regulators due to ageing. Surprisingly, the regulatory influence of ageing on pneumocytes type II cells showed noticeable similarity with patterns due to the effect of novel coronavirus infection in human lung.

Complex Networks, Gene Expression and Cancer Complexity: A Brief Review of Methodology and Applications

In this brief survey, various aspects of cancer complexity and how this complexity can be confronted using modern complex networks’ theory and gene expression datasets, are described. In particular, the causes and the basic features of cancer complexity, as well as the challenges it brought are underlined, while the importance of gene expression data in cancer research and in reverse engineering of gene co-expression networks is highlighted. In addition, an introduction to the corresponding theoretical and mathematical framework of graph theory and complex networks is provided. The basics of network reconstruction along with the limitations of gene network inference, the enrichment and survival analysis, evolution, robustness-resilience and cascades in complex networks, are described. Finally, an indicative and suggestive example of a cancer gene co-expression network inference and analysis is given.

Improving gene-network inference with graph-wavelets and making insights about ageing associated regulatory changes in lungs

Using gene-regulatory-networks based approach for single-cell expression profiles can reveal un-precedented details about the effects of external and internal factors. However, noise and batch effect in sparse single-cell expression profiles can hamper correct estimation of dependencies among genes and regulatory changes. Here we devise a conceptually different method using graph-wavelet filters for improving gene-network (GWNet) based analysis of the transcriptome. Our approach improved the performance of several gene-network inference methods. Most Importantly, GWNet improved consistency in the prediction of generegulatory-network using single-cell transcriptome even in presence of batch effect. Consistency of predicted gene-network enabled reliable estimates of changes in the influence of genes not highlighted by differential-expression analysis. Applying GWNet on the single-cell transcriptome profile of lung cells, revealed biologically-relevant changes in the influence of pathways and master-regulators due to ageing. Surprisingly, the regulatory influence of ageing on pneumocytes type II cells showed noticeable similarity with patterns due to effect of novel coronavirus infection in Human Lung.

Transcriptomic time-series analysis of cold- and heat-shock response in psychrotrophic lactic acid bacteria

Background: Psychrotrophic lactic acid bacteria (LAB) species are the dominant species in microbiota of cold-stored modified-atmosphere-packaged food products and they are the main cause of food spoilage. But still, the cold- and heat-shock response of the spoilage-related psychrotrophic lactic acid bacteria has not been studied. Here, to study cold- and heat-shock response of spoilage lactic acid bacteria, we performed time-series RNA-seq for Le. gelidum, Lc. piscium and P. oligofermentans using temperatures of 0 °C, 4 °C, 14 °C, 25 °C and 28 °C. Results: We showed that the cold-shock protein A (cspA) gene was the main cold-shock protein gene among cold-shock protein genes in all three species. Our results indicated DEAD-box RNA helicase genes (cshA, cshB) play a critical role in cold-shock response in psychrotrophic LAB. In addition, several RNase genes were also involved in cold-shock response in Lc. piscium and P. oligofermentans. Moreover, gene network inference analysis provided candidate genes involved in cold-shock response. Ribosomal proteins, tRNA modification, rRNA modification, and ABC and efflux MFS transporter genes clustered with cold-shock response genes in all three species, which was a strong indication that these genes would be part of cold-shock response machinery. Heat-shock treatment caused upregulation of Clp protease and chaperone genes in all three species and we were able to identify transcription binding site motifs for heat-shock response genes in Le. gelidum and Lc. piscium. Finally, we showed that food spoilage-related genes were upregulated at cold temperatures. Conclusions: The results of this study provide new insights into a better understanding of the cold- and heat-shock response in psychrotrophic LAB. In addition, candidate genes involved in cold- and heat-shock response predicted using gene network inference analysis could be used as a target for future studies.

corto: a lightweight R package for gene network inference and master regulator analysis

Motivation Gene network inference and master regulator analysis (MRA) have been widely adopted to define specific transcriptional perturbations from gene expression signatures. Several tools exist to perform such analyses but most require a computer cluster or large amounts of RAM to be executed. Results We developed corto, a fast and lightweight R package to infer gene networks and perform MRA from gene expression data, with optional corrections for copy-number variations and able to run on signatures generated from RNA-Seq or ATAC-Seq data. We extensively benchmarked it to infer context-specific gene networks in 39 human tumor and 27 normal tissue datasets. Availability and implementation Cross-platform and multi-threaded R package on CRAN (stable version) https://cran.r-project.org/package=corto and Github (development release) https://github.com/federicogiorgi/corto. Supplementary information Supplementary data are available at Bioinformatics online.

corto: a lightweight R package for Gene Network Inference and Master Regulator Analysis

MotivationGene Network Inference and Master Regulator Analysis (MRA) have been widely adopted to define specific transcriptional perturbations from gene expression signatures. Several tools exist to perform such analyses, but most require a computer cluster or large amounts of RAM to be executed.ResultsWe developed corto, a fast and lightweight R package to infer gene networks and perform MRA from gene expression data, with optional corrections for Copy Number Variations (CNVs) and able to run on signatures generated from RNA-Seq or ATAC-Seq data. We extensively benchmarked it to infer context-specific gene networks in 39 human tumor and 27 normal tissue datasets.AvailabilityCross-platform and multi-threaded R package on CRAN (stable version) https://cran.rproject.org/package=corto and Github (development release) https://github.com/federicogiorgi/[email protected]