De novo 3D models of SARS-CoV-2 RNA elements and small-molecule-binding RNAs to guide drug discovery

Ramya Rangan; Andrew M. Watkins; Wipapat Kladwang; Rhiju Das

For each stem-loop in the 5′ UTR and initial stretch of the ORF1a/b coding region, we generated at least 200,000 FARFAR2 models. In Fig. 1B -G, we depict clusters for each 5′ UTR stem-loop from simulations that produced multiple-occupancy clusters, including SL1, SL2, SL3, SL4ab, SL6, and SL7. As expected for these smaller RNA segments, all these stem-loops had excellent modeling convergence. Most clusters had occupancies of greater than 1, indicating that numerous independent de novo modeling trajectories resulted in conformations similar to within 2 Å RMSD. Furthermore, the mean pairwise RMSD of 10 lowest energy cluster centers (Table 1) approached 5 Å or better, suggesting that if a single dominant structure exists for these elements, our average model accuracy would be around 8 Å RMSD or better. Nevertheless, SL1-3 have four or more clusters with E-gap values less than 1 R.E.U., suggesting the presence of many distinct structures at a 2 Å clustering radius that may be trapped by a small molecule without substantial energetic penalty.