Convolutional neural network for automated peak detection in reversed-phase liquid chromatography

Area of Science:

• Analytical Chemistry
• Computational Chemistry

Background:

• Manual inspection and rule-based algorithms for peak detection in chromatography are labor-intensive and can be error-prone.
• Existing automated methods often require significant parameter tuning to balance true and false positive rates.

Purpose of the Study:

• To develop and evaluate a deep learning model for automated peak detection in reversed-phase liquid chromatography (RPLC).
• To assess the performance of a convolutional neural network (CNN) trained on simulated data for peak identification.

Main Methods:

• Implementation of a convolutional neural network (CNN) model to process entire chromatograms.
• Training the CNN model on a large dataset of one million simulated chromatograms.
• Validation of the model using simulated and experimental RPLC chromatograms, comparing against a Savitzky-Golay derivative-based approach.

Main Results:

• The CNN model achieved high performance on simulated data (ROC-AUC of 0.996).
• The model demonstrated comparable or superior performance to the Savitzky-Golay algorithm on experimental data, with an 8.6% increase in true positives.
• Predicted peak probabilities provide a measure of the CNN's confidence in detected peaks.

Conclusions:

• Deep learning, specifically CNNs trained on simulated data, offers an effective approach for automated peak detection in RPLC.
• This method reduces the need for manual data labeling and correction.
• Potential limitations include ensuring the simulated data adequately represents real-world chromatographic complexity.