RVAgene: generative modeling of gene expression time series data

Area of Science:

• Computational biology
• Genomics
• Machine learning in biology

Background:

• Existing methods for modeling genome-wide gene expression dynamics are insufficient for large, temporally rich, or single-cell datasets.
• Variational autoencoders are effective for characterizing large datasets, including single-cell data.
• This study extends variational autoencoder methods for gene expression time series analysis.

Purpose of the Study:

• To develop and present RVAgene, a novel recurrent variational autoencoder for modeling gene expression dynamics.
• To enable accurate and efficient reconstruction of temporal gene profiles from large datasets.
• To facilitate biological feature discovery and generation of new gene expression data through latent space representation.

Main Methods:

• Implementation of a recurrent variational autoencoder (RVAgene) for gene expression time series.
• Utilizing a recurrent encoder network to learn a low-dimensional latent space representation.
• Testing RVAgene on simulated and real biological datasets, including embryonic stem cell differentiation and kidney injury response.
• Applying clustering and gene ontology term enrichment analysis to the learned latent space.

Main Results:

• RVAgene accurately and efficiently reconstructs complex temporal gene expression profiles.
• A low-error latent space representation can be learned using a fraction of the data.
• Unsupervised discovery of biological features, including new gene regulation programs for Lox family genes in kidney injury response.
• Successful application to diverse datasets like embryonic stem cell differentiation and kidney injury.

Conclusions:

• RVAgene provides a powerful new method for analyzing gene expression dynamics in large, time-series datasets.
• The model enables accurate reconstruction, efficient feature discovery, and data generation.
• RVAgene demonstrates potential for uncovering novel biological insights, such as specific gene regulatory mechanisms.