GARN3: A coarse-grained helix centered technique for RNA 3D structures prediction

The study of the prediction of three-dimensional structures of RNA (ribonucleic acids) has been increasing over the last decades, especially with advances in artificial intelligence. However, there are still many gaps. Among the known techniques, GARN (Game Algorithms for RNa 3D sampling) has demonstrated good results in molecules with long structures containing hundreds of millions of nucleotides. Nevertheless, the GARN technique also left room for improvements when considering its final 3D structure for predicted molecules, which can be more refined by presenting more elements, also known as pseudoatoms. In this study, we improved the last version of the technique GARN, which is GARN2, enhancing the visualization of the 3D model by adding more pseudoatoms to the helices in our technique GARN3. In GARN3, we also added a machine learning prediction model to improve the scoring calculation, aiming to improve the prediction of new molecules. In our tests, we demonstrated good results when comparing GARN3 with other techniques in the literature. GARN3, in the same way as GARN2, presents better results for large molecules. In our tests, the simulations with GARN3 demonstrated good results, where GARN3 predicted the majority of molecules (approximately 80% of the test set) better than the older versions of GARN.