Marko LudaicPhD Student

RNA Structure Prediction: Promising Starts, But a Long Road Ahead

Deep learning, a form of artificial intelligence inspired by how the human brain processes information, is now making waves in the molecular sciences. One of the important challenges is predicting the structure of diverse molecules, such as proteins where AI has made tremendous advances, thanks to AI-driven tools like AlphaFold. However, when it comes to RNA, progress has been much slower. RNA plays a crucial role in the cell, from regulating the activity of genes during development to helping cells respond to changes in their environment, yet predicting its 3D structure remains challenging. We tested several deep learning methods for predicting the structure of single RNAs, as well as RNA-RNA and RNA-protein complexes. The results show that simpler RNAs with a classic three-leaf clover shape such as transfer RNA (tRNA), a small RNA molecule that plays a key role in protein synthesis, are predicted more accurately than complex RNAs with intricate folds. Interestingly, the methods did a decent job at identifying where RNAs interact with proteins, unless the RNA was much smaller or the protein was symmetrical, the latter causing the RNA to end up on the wrong side of the symmetrical protein. Overall, while current AI tools show moderate capability for predicting RNA structure, they are challenged with complex or novel forms of the structure, which highlights the need for continued research to tackle the full RNA diversity found in nature.