Structure primed embedding on the transcription factor manifold enables transparent model architectures for gene regulatory network and latent activity inference

ABSTRACT: Regulation of gene expression in biological systems is a complex, nonlinear process composed of context specific interactions, from signaling and transcription to genome modification. Modeling gene regulatory networks (GRNs) can be limited due to a lack of direct measurements of regulatory features in genome-wide screens. Most GRN inference methods are consequently forced to model covariance between regulatory genes and their targets as a proxy for causal interactions. This in turn complicates validation and reuse of predictive modeling frameworks. To disentangle covariance and casual influence require aggregation of independent and complementary sets of evidences, such as transcription factor (TF) binding and target gene expression. Common approaches include the overlap of evidence to infer causal relations. However the complete state of the system, e.g. TF activity (TFA) is unknown. Other methods tries to estimate these latent features. These models often use linear frameworks that are unable to account for non-linearities, TF-TF interactions, and other higher order features. Deep learning frameworks can be used to model complex interactions between features and capture latent features of higher order. However deep learning methods often discard central concepts in biological systems modeling such as sparsity and latent feature interpretability in favour of increased complexity of the model. In this work we demonstrate that gene regulatory network inference using latent features such as transcription factor activity can be built into a single framework. We present a novel deep learning approach (the Supirfactor framework) that incorporates multiple data-type orthogonal evidence of regulation and maintains interpretable parameter estimates.

ORGANISM(S): Saccharomyces cerevisiae

PROVIDER: GSE218089 | GEO | 2023/01/31

REPOSITORIES: GEO