Chromatogram Peak Analysis for Multimodal and Overlap Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional peak detection methods using machine learning struggle to accurately distinguish multimodal peaks from overlapping peaks in chromatograms, particularly at low component concentrations, leading to false identifications of trough portions as peak-ending or peak-beginning points.
Innovation Solution
A waveform-analyzing method that creates a trained model using partial waveforms to identify peak regions, including single-peak, overlap-peak, and non-peak regions, and determines whether an overlap peak is multimodal by analyzing peak height, trough depth, and horizontal axis width, allowing for accurate classification and correction of peak detection results without manual intervention.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If machine learning is used for peak detection in chromatograms, then peak detection automation and efficiency are improved, but false identification of multimodal peaks as overlapping peaks increases
Solution Approach 1:
The chromatogram is divided into multiple partial waveforms, each processed independently by the trained model to identify peak regions. This segmentation allows the system to handle complex multimodal peaks by analyzing smaller segments and integrating results, reducing false identifications while maintaining automation.
Solution Approach 2:
A trained model is created in advance using machine learning on reference waveforms with known peak positions. This pre-trained model captures the characteristics of multimodal peaks, enabling accurate identification during actual analysis without manual intervention, thus resolving the contradiction between automation and accuracy.
2Measurement precision
If conventional peak detection methods are used, then simple peaks are detected accurately, but multimodal peaks at low concentrations are misidentified as overlapping peaks
Solution Approach 1:
The system changes the approach from conventional fixed-threshold methods to machine learning-based parameter analysis. The trained model learns optimal parameters for identifying multimodal peaks by analyzing reference waveforms, enabling accurate detection across varying concentrations and peak shapes without manual adjustment.
Solution Approach 2:
The system creates a trained model that copies the characteristics of known peak patterns from reference waveforms. This model is then applied to analyze unknown peaks, allowing the system to recognize multimodal patterns without manual intervention and improving both precision and adaptability.
3Speed
If machine learning models are trained on reference waveforms, then peak detection speed is improved, but the model may overfit to specific peak shapes and fail on varied multimodal peaks
Solution Approach 1:
The model processes chromatograms by dividing them into partial waveforms and analyzing each segment. This segmentation approach allows the model to maintain speed while adapting to varied peak shapes, as each segment can be independently evaluated without requiring the entire chromatogram to match a specific pattern.
Solution Approach 2:
The trained model is designed to be universal, capable of handling multiple types of peaks including single peaks, multimodal peaks, and overlapping peaks. By training on diverse reference waveforms and using region estimation to identify different peak types, the model achieves both speed and generalization capability.
Data Source
AI summary
A model for locating a peak portion in a chromatogram or similar waveform is trained by machine learning using multiple sets of partial waveforms prepared by dividing each reference waveform having a known peak position. An analysis-target waveform is divided into partial waveforms, and whether or not a partial waveform is a peak portion is determined for each partial waveform by the trained model, to estimate different kinds of regions in the analysis-target waveform, including an overlap-peak region. For an overlap peak within a region estimated to be an overlap-peak region, whether or not that peak is a multimodal peak originating from one component is determined, using at least the height of a peak in the overlap peak, the depth of the trough between two neighboring peaks, or the horizontal width of the portion between the bottom of the trough and the top of one of the neighboring peaks.