ENTRNA: a framework to predict RNA foldability

Area of Science:

• Computational Biology
• Bioinformatics
• Molecular Biology

Background:

• RNA molecules are vital for cellular functions, with their spatial structures dictating biological roles.
• Understanding RNA folding, especially secondary structures, is crucial for deciphering these functions.
• Current RNA design methods often treat it as a structure prediction or inverse folding problem, heavily relying on free energy calculations.

Purpose of the Study:

• To develop a novel data-driven framework, ENTRNA, for predicting RNA foldability.
• To address limitations in existing RNA design methodologies by reframing it as a foldability prediction problem.
• To introduce a new feature, Sequence Segment Entropy (SSE), for measuring RNA sequence diversity.

Main Methods:

• Developed the ENTRNA framework using a Positive-Unlabeled learning approach.
• Incorporated Sequence Segment Entropy (SSE) alongside traditional sequence and structural features.
• Trained and validated ENTRNA on extensive datasets from the RNASTRAND database (1024 pseudoknot-free, 1060 pseudoknotted RNAs).
• Tested model robustness on independent datasets from the PDB (206 pseudoknot-free, 93 pseudoknotted RNAs).

Main Results:

• ENTRNA achieved 86.5% sensitivity on the pseudoknot-free training set and 80.6% on the testing set.
• For pseudoknotted RNAs, ENTRNA demonstrated 81.5% sensitivity on the training set and 71.0% on the testing set.
• Successfully predicted the foldability of 4 out of 5 long, structurally complex synthetic RNAs.

Conclusions:

• Reformulated RNA design as a foldability prediction problem, estimating the likelihood of sequence-structure pair co-existence.
• The ENTRNA framework offers potential advancements for both RNA structure prediction and inverse folding problems.
• This work highlights the utility of data-driven approaches in RNA research.