Prediction of DNA i-motifs via machine learning

Area of Science:

• Genomics
• Bioinformatics
• Molecular Biology

Background:

• i-Motifs (iMs) are DNA secondary structures crucial for genomic functions.
• While common in the human genome, their folding status and stability vary significantly.
• Existing research relies on biophysical experiments, lacking dedicated predictive computational tools.

Purpose of the Study:

• To introduce iM-Seeker, a novel machine learning pipeline for predicting DNA iM folding status and structural stability.
• To provide a computational solution for analyzing iM structures genome-wide.

Main Methods:

• Utilized a Balanced Random Forest classifier for predicting iM folding status using CUT&Tag sequencing data.
• Employed an Extreme Gradient Boosting regressor to estimate iM folding strength, integrating literature and experimental biophysical data.
• Trained and validated the models on extensive datasets.

Main Results:

• iM-Seeker achieved 81% accuracy in predicting DNA iM folding status.
• The model estimated folding strength with a coefficient of determination (R2) of 0.642 on the test set.
• Nucleotide composition analysis revealed cytosine and thymine positively correlate with iM stability, while guanine and adenine show a negative correlation.

Conclusions:

• iM-Seeker offers a reliable computational method for predicting DNA i-motif folding status and stability.
• The tool enhances the understanding of iM formation and its genomic implications.
• Sequence composition is a key determinant of iM stability, guiding future research and applications.