Abstract

Timetable

Download Conference Program

Download All Abstracts

Zoom Access

Plenary Lectures Day1　Day2　Day3　Day4

MSSJ Award Lectures Day1

Oral Sessions Day2　Day3　Day4

Poster Presentations Day1　Day2　Day3　Day4

Late-Breaking Posters Day1

Luncheon Seminars Day2　Day3　Day4

Teatime Sessions Day3

Corporate Posters Day2　Day3　Day4

Corporate Program

Poster Presentations

Day 2, June 23（Mon.）　

Room P (Maesato East, Foyer, Ocean Wing)

• 2P-AM-10
• PDF

Development of Global Machine Learning Models for Understanding Retention Mechanisms and Predicting Retention Time in Supercritical Fluid Chromatography/Mass Spectrometry

(¹Kyushu Univ., ²Div. of Metabolomics, MIB, Kyushu Univ.)
^oOmidreza Heravizadeh¹, Kohta Nakatani^1,2, Noriyuki Tomiyasu², Taihei Torigoe², Toshiyuki Yamashita², Masatomo Takahashi^1,2, Yoshihiro Izumi^1,2, Takeshi Bamba^1,2

Supercritical fluid chromatography (SFC) is a high-throughput separation technique offering rapid analysis, enhanced efficiency, and superior sensitivity when coupled with mass spectrometry (MS). However, its complex retention mechanism, influenced by variations in mobile phase density and solvation power, remains challenging to understand completely. Predicting retention time (RT) accurately is essential for efficient method development and reliable compound identification in SFC/MS analysis.
In this study, we developed two global machine learning-based quantitative structure-retention relationship (QSRR) models using molecular and system descriptors to predict RT across different analytical conditions. SFC/MS analysis of over 1200 compounds was performed using polar and non-polar stationary phases with multiple mobile phase compositions. The models, trained with optimized hyperparameters, achieved high predictive performance (R² = 0.89 and 0.85 for polar and non-polar phases, respectively).
Partial least squares regression (PLS-R) analysis revealed that molecular descriptors related to spatial distribution characteristics and binding ability significantly influence retention in both polar and non-polar phases. Additionally, system descriptors, representing stationary and mobile phase properties, played a crucial role in analyte retention in non-polar stationary phases. These insights improve RT prediction accuracy and deepen our understanding of SFC retention, facilitating method development and confident compound identification.

---