Multilevel ensemble for genome-wide prediction of cell-specific G-quadruplexes utilizing transformed sequences and enhanced chromatin accessibility with G4Beacon2.

G4Beacon2 achieves accurate genome-wide in vivo cell-specific G-quadruplex prediction with a multilevel ensemble model, outperforming existing tools across multiple cell lines.

• Why it matters: Understanding cell-specific G4 structures is vital for insights into their biological roles, yet current computational methods lack genome-wide, in vivo, cell-specific prediction capabilities, limiting research progress.
• What they did: The model employs DNABERT2 for DNA sequence semantic analysis, normalizes chromatin accessibility with Z-score, and integrates a three-level ensemble framework, tested across intra- and cross-cell-line datasets with multisource data.
• The result: G4Beacon2 demonstrates consistent high performance, with evidence that mouse training data can enhance human predictions, providing a robust, user-friendly tool for genome-wide G4 analysis in various cell types.