SOORENA: Self-lOOp containing or autoREgulatory Nodes in biological network Analysis

Autoregulatory mechanisms, in which proteins regulate their own activity or expression, are fundamental to biological networks but are challenging to identify systematically from literature. To address this gap, we present SOORENA (https://soorena.it.helsinki.fi/soorena/), a two-stage transformer model that predicts and classifies protein autoregulation in PubMed abstracts. SOORENA was trained on 1,332 experimentally validated abstracts and achieved 96.0 percent accuracy and 97.8 percent precision in stage one, with stage two achieving 95.5 percent accuracy and 96.2 percent macro-F1 across seven mechanistic classes. Applied to 3.34 million abstracts, SOORENA identified 85,145 publications containing autoregulatory mechanisms, yielding 97,657 protein-specific records. Integration with curated databases generated 100,065 comprehensive entries accessible via an interactive Shiny application. By systematically cataloging self-regulatory interactions, which often act as bottlenecks in dynamic network modeling, SOORENA provides a resource that supports mechanistic interpretation, model reduction, and predictive systems-level analyses. These results demonstrate that domain-specific language models can scale the discovery and curation of biologically essential self-regulatory mechanisms, bridging literature mining and systems biology.