Authors:

Nicholas Paredes

Date Created:

Not specified

Course Title:

Professor:

Not specified

About Paper:

G-quadruplexes (G4s) are DNA secondary structures formed in guanine-rich genomic regions. G4s are enriched in human telomeres and oncogene promoters, where they play important roles in genomic stability and transcription. G4s are therefore attractive targets for anticancer drugs. To find new genomic G4s, software has been developed to predict if DNA sequences can form G4s, which can then be experimentally assessed. However, it is unclear which software is best, regarding performance and usability. Herein, I compared three different software for their predictive capabilities: pqsfinder, G4Hunter, and G4mismatch. I tested software with a publicly available dataset of 392 experimentally validated sequences, including 298 G4-forming sequences (positives) and 94 non-G4-forming sequences (negatives). Using each software's default settings, I determined their accuracy, precision, recall, and false discovery rate (FDR). G4Hunter has the highest overall accuracy and recall but the lowest precision. pqsfinder has the highest precision but the lowest accuracy and recall. G4mismatch has intermediate accuracy, precision, and recall. All FDRs are <5%. Predictive scores are significantly greater (P<0.001) for G4-forming sequences than non-G4-forming sequences for all software. All false positives have above-average guanine content, indicating that guanine overabundance in non-G4- forming sequences can lead to false positive predictions. The percentages of false negatives containing either double-guanine tracts or long nucleotide loops are >80% across all software, meaning that these features may lead to false negative predictions. G4Hunter and G4mismatch have better usability with large datasets than pqsfinder. Based on its accuracy and recall, as well as its usability, G4Hunter outperforms the other software. Keywords: G-quadruplex; Bioinformatics; Structural Prediction

Source:

Purdue University / 2024

Topics:

No topics listed

Co-authors:

Nicholas Paredes