PhD Scientific Days 2025

Poster Session III. - H: Pharmaceutical Sciences and Health Technologies

Machine Learning-Assisted Retention Time Predictions on Polysaccharide-based Chiral Columns

Name of the presenter

Dombi Gergely

Institute/workplace of the presenter

Semmelweis University Department of Pharmaceutical Chemistry

Authors

Gergely Dombi¹, Attila Imre², Máté Dobó¹, Ferencz Elek³, Balázs Balogh⁴, Anna Vincze¹, Zoltán-istván Szabó⁵, György Tibor Balogh¹, Anita Rácz⁶, Gergő Tóth¹

1: Semmelweis University Department of Pharmaceutical Chemistry
2: Semmelweis University Center for Health Technology Assessment
3: Emergency County Hospital Miercurea Ciuc
4: Semmelweis University Department of Organic Chemistry
5: George Emil Palade University of Medicine, Pharmacy, Science and Technology of Targu Mures Department of Pharmaceutical Industry and Management
6: HUN-REN Research Centre for Natural Sciences Institute of Materials and Environmental Chemistry

Text of the abstract

Introduction: Enantioseparation presents a significant challenge in analytical chemistry. This process often requires numerous trials with various experimental conditions using high-performance liquid chromatography (HPLC).
Aims: We propose a solution that uses machine learning techniques. Our approach utilizes consensus modeling based on the Partial Least Squares (PLS) regression method, neural network (NN) algorithms, and a graph neural network (GNN) method. This combination aims to predict the retention times of compounds on polysaccharide-based chiral stationary phases under different polar organic mode mobile phases.
Method: For the development of the machine learning methods, we compiled a homogeneous dataset of over 500 unique molecules and more than 1300 retention time measurements under four polar organic mode conditions (acidic and basic methanol, acidic and basic acetonitrile).
Results: The PLS-NN consensus model excelled in condition-specific predictions, achieving R² values above 0.7 and RMSECV values below 0.4 in most scenarios. Conversely, the GNN model demonstrated superior performance in combined predictions across all conditions, achieving an R² of 0.84 and RMSECV of 0.38 for cross-validation, and an R² of 0.84 and RMSEP of 0.23 on the test set. The GNN model was implemented to create a freely accessible webserver at chiralscreen.com.
Conclusion: Our work introduces an innovative approach for predicting chiral separations, offering a freely accessible web application to the scientific community. The tool is designed to provide accurate and reliable predictions, enhancing the efficiency and effectiveness of enantioseparation processes.
In summary, the integration of machine learning methods presents a significant advancement in chiral analyses. By providing a reliable and efficient prediction tool, we aim to support researchers in achieving faster method developments for chiral molecules.
Funding: Supported by the EKÖP-2024-57 New National Excellence Program of the Ministry for Culture and Innovation from the Source of the National Research, Development and Innovation Fund. This work was funded by the National Research, Development, and Innovation Office Hungary. This work was supported by the János Bolyai Research Scholarship of the Hungarian Academy of Sciences