Marcell Szikszai

This paper evaluates deep learning methods for RNA secondary structure prediction, highlights generalization challenges, and discusses integration with traditional methods.

RNAcentral integrates literature and gene-level data to enhance non-coding RNA sequence analysis and functional understanding.

The paper identifies differences between protein and RNA structure and proposes RNA-specific methods to improve deep learning models for RNA 3D structure prediction.

The paper introduces RNA3DB, a dataset of structured RNAs designed for training and benchmarking deep learning models, with non-redundant sequence and structure splits to improve generalization.

The paper shows that deep learning models for RNA secondary structure prediction do not generalize across families, using rigorous cross-validation methods.