Machine Learning for Group-Targeted Elution Order Prediction: Substituted Flavones as a Case Study.

Area of Science:

• Analytical Chemistry
• Computational Chemistry
• Cheminformatics

Background:

• Accurate prediction of elution order for plant metabolites is crucial for dereplication.
• Machine learning (ML) offers an efficient method for automating peak annotation using structure-retention relationships.

Purpose of the Study:

• To train and validate ML models for predicting the elution order of flavonoid derivative pairs.
• To explore the impact of molecular fingerprints and training strategies on prediction accuracy.

Main Methods:

• Four ML models (ranking neural networks, logistic regression) were trained using a novel molecular fingerprint.
• Two datasets (51 compounds, 48 compounds) were constructed from literature data.
• Retention times were indirectly estimated using linear regression and interpolation.

Main Results:

• Pairwise error rates for elution order prediction were consistently below 10% under reversed-phase LC conditions.
• Linear regression models showed a slight performance advantage, statistically validated.
• Larger, uniform datasets proved more beneficial than fragmented ones for model training.

Conclusions:

• ML models, particularly linear regression, demonstrate reliable performance in predicting elution order for flavone derivatives.
• The developed molecular fingerprint effectively encodes structural features for phytochemical analysis.
• Understanding substituent effects on chromatographic retention is enhanced through model weight visualization.