Improving the folding prediction of RNA with deep learning

L.A. BUGNON, L. DI PERSIA, M. GERARD, A. EDERA, J. RAAD, S. PROCHETTO, E. FENOY, G. STEGMAYER AND D.H. MILONE.

Conferencia; XIICAB2C - 12th Argentinian Conference in Bioinformatics and Computational Biology; 2022

Thefunction of noncoding RNAs (ncRNAs) is relevant for numerous biologicalprocesses and largely depends on their secondary structure, which determinesinteractions with partner molecules. However, the determination of ncRNA structuresis a very costly process, which cannot be scaled up efficiently, limiting ourability to functionally characterize such molecules. Computational methods arepromising for  the prediction of ncRNAstructures, which is speeding up the discovery of function and actionmechanisms. By contrast, classical tools strongly rely on hand-craftedthermodynamic features with limited capacity for modeling the wide structuraldiversity of RNAs, including pseudoknots, non-canonical bonds and longsequences. Recently, machine learning techniques have demonstrated being ableto capture thermodynamic features in an automated manner