Quantitative Structure Retention Relationship Models in an Analytical Quality by Design Framework: Simultaneously Accounting for Compound Properties, Mobile-Phase Conditions, and Stationary-Phase Prop

详细信息    查看全文

• 作者：Koji Muteki ; James E. Morgado ; George L. Reid ; Jian Wang ; Gang Xue ; Frank W. Riley ; Jeffrey W. Harwood ; David T. Fortin ; Ian J. Miller
• 刊名：Industrial & Engineering Chemistry Research
• 出版年：2013
• 出版时间：September 4, 2013
• 年：2013
• 卷：52
• 期：35
• 页码：12269-12284
• 全文大小：972K
• 年卷期：v.52,no.35(September 4, 2013)
• ISSN：1520-5045

文摘

Quantitative structure retention relationships (QSRRs) can play an important role in enhancing the speed and quality of chromatographic method development. This paper presents a novel (compound-classification-based) QSRR modeling strategy that simultaneously accounts for the analyte properties, mobile-phase conditions, and stationary-phase properties. It involves the adoption of two models: (A) partial-least-squares discriminate analysis (PLS-DA) to classify compounds into subclasses having similar interactive relationships between the mobile-phase conditions and stationary phase; (B) L partial least squares (L-PLS) to predict the compound鈥檚 retention time based on the mobile-phase conditions, stationary phase, and compound properties. For the retention time of a compound to be modeled, the most favorable compound class is identified in an optimization framework that simultaneously minimizes both the compound misclassification rate (based on PLS-DA) and the retention time prediction error (based on L-PLS) through a mixed-integer optimization. The proposed QSRR model (L-PLS with compound classification) significantly improves the retention time predictability compared with traditional QSRR or L-PLS models without compound classification. When combined with the linear solvation energy relationship parameters (using Abraham coefficients) as the column properties, the approach allows the following: (1) prediction of (new, never analyzed) compound retention times under chromatographic conditions (columns and mobile-phase conditions) used to train the model; (2) prediction of (previously analyzed under training conditions) compound retention times under chromatographic conditions that have not been previously evaluated; (3) optimization of the chromatographic conditions (mobile-phase and column selection) to maximize critical pair resolution, including new compounds; (4) enhanced mechanistic understanding of the interactive retention relationship between compounds, the mobile phase, and the column (e.g., compound retention mechanism). The effectiveness of the proposed modeling strategy will be demonstrated through two practical pharmaceutical applications in supercritical fluid chromatography and reversed-phase liquid chromatography.