Comparison of different de novo modelling approaches for the FRET prediction of structural ensembles of ribonucleic acids.
Mirko Weber, Felix Erichson, Richard Börner
The integrative modelling of biomolecules relies on both a structural ensemble of a biomolecule and reliable experimental data reporting on its binding and folding trajectories. Therein, structural ensembles can be generated by de novo modelling tools or by all-atom molecular dynamics (MD) simulations. However, MD simulations are computationally expensive and limited regarding sampling capabilities of the conformational biomolecular ensemble [1]. We therefore use three different tools, RNAcomposer [2], Rosetta's FARFAR2 [3] and Alphafold3 [4], for the de novo RNA 3D structure prediction of a ribosomal RNA tertiary contact comprising a kissing loop and a GAAA tetraloop motif, generating a highly diverse structural ensemble. Subsequently we compute the accessible contact volume (ACV) of the FRET pair sCy3 and sCy5 using FRETraj [5] to predict the FRET distribution of each structural ensemble. We use the experimental FRET distribution to weight and analyse the predicted FRET distributions based on the structural ensemble of all tools. Our results indicate that RNAcomposer and Alphafold3 capture only a limited subset of possible structures and cannot replicate the full range of the experimental low FRET unbound state. In contrast, FARFAR2 generates a highly diverse structural ensemble that encompasses a wide range of possible structural conformations and almost perfectly replicates the experimental smFRET distribution.