Generation of 3D models of miRNA implicated in interactions with NCL
Structural information of miRNA molecules implicated to play a role in breast cancer (Supporting Table S4 ) and interact with NCL is unavailable in the databases. The limited structural data available in the protein databank for the 6 miRNA molecules under study corresponds to either partial pre-miRNA structures or apical loops (PDB ID: 2MNC, 5UZT) [77,78] and neither provide structural information for the complete pri-miRNA structures. Therefore, all pri-miRNA molecules were modeled using the primary sequences of miRNA stem loop structures obtained from miRbase [79]. The secondary structure of these miRNA molecules was predicted using various programs including RNAStructure [80], RNAFold [81], MC-Fold [82], CentroidFold [83], and SPOT-RNA[84]. Although all these tools provide predictions based on experimental data found in databases, some of the newer methods such as CentroidFold and SPOT-RNA also utilize machine learning in making these predictions. All of the predicted secondary structures (dot bracket format) obtained from these tools were input into the RNA modeling software including RNAComposer [85], ifold2 [86], RNAfold3 [87], 3DRNA v2 [88] and simRNAweb [89] to obtain 3D models of these miRNA molecules. The generated models were evaluated for their structural quality using MolProbity [90] (Supporting Table S3 ). The top models were refined using RNAfitme [91], a program that reconstructs nucleobases and nucleosides while keeping the sugar phosphate backbone fixed and aiming to reduce steric clashes.