Discovering a Novel Regulatory Motif in Innate Immunity with Deep Learning

Signaling processes dominate cellular responses to stimuli and are integral to understanding biological processes.

Yet most experimental methods are ill-equipped to measure signaling activity, while current analyses are plagued by incomplete and biased annotations.

Here, we will develop and apply DeepSignaling, a multi-view deep learning approach that assesses systems biology data without reference to any annotation and solely with respect to its relevance for explicitly measured signaling activity.

This allows us to identify novel components of signaling pathways without the bias of canonical pathways and annotated gene sets.

A preliminary version of DeepSignaling, applied to macrophage signaling, identified candidate key components of a novel negative feedback loop regulating immune cell activity.

We will experimentally validate this new signaling element, revolving around catechol-O-methyltransferase (COMT), in human macrophages and other immune cell types, such as T helper cells, using immune cell co-cultures, liquid chromatography-tandem mass spectrometry, and cytokine assays.

This will be important for understanding aberrant macrophage activity in disease contexts, such as pre-eclampsia and cancer.

We will further advance and generalize DeepSignaling to retrieve signaling predictions in more biological contexts, to establish a platform for expanding our understanding of signaling pathways and provide further insight into their role in disease.