# Computational understanding of non-coding RNA pairwise interactions

## TL;DR

This paper introduces CUPID, a deep learning tool that predicts interactions between non-coding RNAs using only their sequence data.

## Contribution

CUPID is the first deep learning framework that generalizes across various types of non-coding RNAs to predict pairwise interactions.

## Key findings

- CUPID uses RNA sequence embeddings and a classifier to predict ncRNA interactions without thermodynamic models.
- The model generalizes across different ncRNA types like long non-coding, circular, and micro-RNAs.
- CUPID provides a scalable approach for studying RNA-based regulatory networks.

## Abstract

Non-coding RNAs (ncRNAs) govern a vast network of regulatory interactions within the cells, yet their pairwise relationships remain largely uncharted due to the complexity of RNA structure and the limits of current experimental methods. We present CUPID (Computational Understanding of Pairwise Interactions in ncRNA Data), a deep learning framework that predicts ncRNA-ncRNA interactions directly from primary sequence information. CUPID uses embeddings from a pre-trained RNA language model combined with a feed-forward classifier to identify patterns linked to molecular pairing. This approach avoids reliance on thermodynamic models or manual feature design and, unlike previously proposed models, is able to generalize across different types of ncRNAs, including long non-coding, circular, micro-, and small nuclear RNAs. By learning the hidden rules that govern RNA recognition, CUPID provides a scalable tool for exploring ncRNA interaction networks and advancing our understanding of RNA-based regulation.