Evaluation of nine bioinformatic platforms using a viral model for RNA secondary structure prediction

Abstract

RNA secondary (2D) structures are essential for cellular processes including protein translation; indeed some RNA viral pathogens use this strategy during replication cycle. Among Flaviviridae family, Hepatitis C Virus (HCV) presents a well experimentally-characterized secondary structure called IRES (Internal Ribosome Entry Site), which could be used as a Viral model for evaluating the accuracy RNA 2D structure in silico tools; the choice of free-available platforms is a principal concern because standard approaches for single-sequence RNA secondary structure prediction are dissimilar, training their parameters with known RNA structures that do not always include viral sequences. Here, we assessed the accuracy of nine web servers using the HCV IRES domain II. The platforms used were classified in three categories according to S and PPV values obtained. The lowest values were observed with IPKNOT 1.2.1 (NUPACK), Kinefold, VsFold 5.23 and CentroidHomfold, while IPKNOT 1.2.1 (CONTRAfold), IPKNOT 1.2.1 (McCaskill) and ContextFold 1.2.1 had acceptable computing output. Finally, MFOLD and RNAfold based on thermodynamic, RNAstructure (MaxExpect) and CentroidFold based on statistical weights, showed the highest accuracy with a similar topology to the expected structure, although these tools did not predict correctly all base-pairings. Further modifications in these platforms will be necessary to improve the RNA 2D structure prediction in viral models and others with higher RNA topology complexities.