Authors:

David Platt

Date Created:

Not specified

Course Title:

Professor:

Not specified

About Paper:

The three-dimensional structure of non-coding RNA is vital in understanding its biological function. Due to the difficult nature of solving structures experimentally, prediction via deep learning offers an effective alternative. Expanding upon Kihara Lab's prediction software, NuFold, the NuFold Database was created with the goal of making RNA tertiary structure predictions as accessible as those for proteins. To create a library of prediction models that accurately depict structural confidence, we implemented methods for evaluating the preservation of secondary structure within the final tertiary model. We quantify this by calculating an F1 score that compares high-confidence secondary structures from prediction methods (IPKnot, an RNA secondary structure prediction method used to run NuFold, and RNACentral's R2DT, a prediction method based on a large library of known RNA structures) to the identified secondary structures in the tertiary model. A high F1 score indicates the fold is successfully maintained in 3D, serving as a metric complementary to local scores like the predicted local distance difference test (pLDDT). We further validate this F1 score by correlating it with metrics relevant to NuFold, primarily multiple sequence alignment depth. to identify statistical relevance. We use these results to provide users with a predicted structure model that both validates its per-nucleotide accuracy, and preservation of secondary structure. Currently the database contains 13,837 individual predicted RNA sequences from the human genome. We plan to extend the data to both model organism and global health noncoding transcriptomes. Keywords: Deep Learning; RNA; Structure Prediction; Tertiary/3D Structure; Database

Source:

Purdue University / 2025

Topics:

No topics listed

Co-authors:

David Platt