Improved workflow for constructing machine learning models: Predicting retention times and peak widths in...

Area of Science:

• Analytical Chemistry
• Biochemistry
• Computational Biology

Background:

• Manual processing of large oligonucleotide datasets is infeasible.
• Accurate prediction of retention times and peak widths is crucial for chromatographic analysis.
• Diverse oligonucleotide forms and gradient slopes require robust analytical methods.

Purpose of the Study:

• To develop an improved workflow for machine learning model development.
• To predict oligonucleotide retention times, peak widths, and peak resolutions.
• To manage and analyze large-scale chromatographic datasets efficiently.

Main Methods:

• Explored native and phosphorothioated oligonucleotides using a C18 chromatographic system.
• Developed a semi-automatic rule-based approach for data processing and analysis.
• Applied machine learning models including Support Vector Regression (SVR), Gradient Boosting (GB), Random Forest (RF), and Decision Tree (DT).

Main Results:

• Gradient Boosting (GB) and SVR models demonstrated strong performance for retention time predictions.
• Machine learning models showed higher errors with shallower gradients and lower predictability for P=O sequences.
• The best models for the dataset were GB and SVR, with potential for predicting impurity peak resolution.

Conclusions:

• The developed workflow enables efficient analysis of large oligonucleotide datasets.
• Predictive models can forecast chromatograms for various gradient slopes and sequences.
• This approach enhances the prediction of impurity peak resolution in oligonucleotide analysis.