SIREN: Predicting RNA-Small Molecule Interaction using Language Models without Requiring Structure
RNA-targeted drug discovery is constrained by the scarcity of resolved RNA tertiary structures and the conformational flexibility of RNA, which limit the applicability of structure-based virtual screening. Here, SIREN (Small molecule Interaction with RNA via sEquence-based predictioN) is presented, a sequence-based framework that ranks small molecules for a given RNA target directly from its primary sequence without requiring tertiary structure. SIREN integrates a pretrained RNA language model with molecular language model embeddings and physicochemical descriptors through an interaction network that outputs a target-specific compound-ranking score. On four held-out ROBIN riboswitch targets, SIREN outperforms the structure-based methods RNAmigos2, RNAmigos++, and rDock in both enrichment and overall discrimination, and a cross-target derangement analysis confirms that its rankings depend on RNA identity rather than general ligand propensity alone. On an independent HIV-1 TAR benchmark, SIREN achieves strong discrimination without an experimentally derived RNA conformational ensemble. Prospective screening against IRES-derived RNA hairpins from cMyc and XIAP mRNAs further identifies multiple experimentally validated, micromolar-affinity binders from a diverse chemical library. These results demonstrate that SIREN generalizes to unseen RNA targets, transfers across independent benchmarks, and provides a scalable, structure-free framework for RNA-targeted compound prioritization that complements existing structure-based screening approaches.
Authors
- Wonkyeong Jang (ORCID: https://orcid.org/0009-0006-4288-9540)
- Woong‐Hee Shin (ORCID: https://orcid.org/0000-0003-3462-0243)
- Jinseo Heo (ORCID: https://orcid.org/0009-0002-7481-9839)
- Yein Kim
- Yeongju Lee
Institutions
- Korea University (KR)
- Pusan National University (KR)
Publication Details
- Journal
- bioRxiv (Cold Spring Harbor Laboratory)
- Published
- 2026-11-20
- DOI
- https://doi.org/10.64898/2026.11.19.768003
- Primary Topic
- Computational Drug Discovery Methods
- Type
- preprint
Funders
- National Research Foundation
- Ministry of Trade, Industry and Energy
- Korea Institute for Advancement of Technology
- Korea Research Institute of Chemical Technology
- National Research Foundation of Korea
- Ministry of Science and ICT, South Korea
- Institute for Information and Communications Technology Promotion