PENGUINN: Precise Exploration of Nuclear G-Quadruplexes Using Interpretable Neural Networks

Area of Science:

• Genomics
• Bioinformatics
• Molecular Biology

Background:

• G-quadruplexes (G4s) are crucial nucleic acid secondary structures involved in genomic functions like DNA replication and transcription.
• Traditional G4 structure models involve four guanine-rich strands with variable loops.
• Experimental G4 identification methods are costly and technically challenging, necessitating computational solutions.

Purpose of the Study:

• To develop an accurate computational method for identifying G-quadruplex (G4) forming sequences.
• To present PENGUINN, a machine learning model designed for G4 prediction.
• To offer accessible tools for evaluating G4 potential in genomic sequences.

Main Methods:

• Utilized Convolutional Neural Networks (CNNs) for machine learning.
• Trained the model on known G4 forming genomic sequences.
• Developed a standalone implementation and a web application for PENGUINN.

Main Results:

• PENGUINN accurately predicts G-quadruplexes (G4s).
• The model outperforms existing state-of-the-art methods in G4 identification.
• Achieved high accuracy in identifying probable G4 forming genomic sequences.

Conclusions:

• PENGUINN offers a powerful and accurate computational approach for G4 identification.
• The developed tools facilitate the study of G4s in genomics.
• Machine learning methods show significant promise for predicting complex nucleic acid structures.