Quantitative sequence-activity model analysis of oligopeptides coupling an improved high-dimension feature selection...

Area of Science:

• * Cheminformatics
• * Bioinformatics
• * Computational Chemistry

Background:

• * Oligopeptide structure-activity relationships are crucial for drug discovery and development.
• * Existing quantitative sequence-activity model (QSAM) methods often face challenges with high-dimensional data and interpretability.
• * Efficient feature selection is critical for building robust QSAM models.

Purpose of the Study:

• * To develop a novel, rapid feature selection method for high-dimensional descriptor data in QSAM.
• * To improve the performance and interpretability of QSAM models for oligopeptides.
• * To establish a more effective tool for regression analysis of complex biological data.

Main Methods:

• * Utilization of 531 physicochemical amino acid property parameters as descriptors for oligopeptides.
• * Development of a Binary Matrix Resetting Filter (BMRF) for nonlinear, high-dimensional feature selection.
• * Application of Multiround Last Elimination (MRLE) for refined feature screening.
• * Construction of regression models using Support Vector Regression (SVR) for QSAM analysis.

Main Results:

• * Demonstrated significant improvement in QSAM modeling performance, particularly in external prediction, compared to existing methods.
• * Achieved enhanced interpretability of the QSAM model by directly linking reserved descriptors to biochemical significance.
• * Validated the novel method on two distinct oligopeptide systems.

Conclusions:

• * The proposed BMRF and MRLE feature selection strategy, combined with SVR, offers a superior approach for QSAM.
• * The method enhances both predictive accuracy and the biochemical interpretability of QSAM models.
• * This novel technique shows high potential as a tool for high-dimensional regression analysis, including peptide and protein QSAM.