User Benefits
Application
News
Peakintelligence peak integration software substantially reduces the time and effort required for data analysis in PFAS analysis.
Since it does not require parameter settings, Peakintelligence allows even inexperienced users to perform peak integration and
achieve results on par with experienced workers.
Efficient Data Analysis with Peakintelligence™
—Application for Analyzing Per- and
Polyfluoroalkyl Substances (PFAS)—
Takanari Hattori and Hiroyuki Niwa
High Performance Liquid Chromatograph Mass Spectrometer LCMS™-8045/8050/8060RX
Introduction
Per- and polyfluoroalkyl substances (PFAS) are lyophobic, heat
resistant, and chemically resistant compounds used extensively as
coatings, surface treatment agents, emulsifiers, and fire-
extinguishing agents. Concerns about PFAS and their persistence
in the environment, accumulation in living organisms, toxicity to
living organisms, and ability to travel long distances have recently
led to fact-finding studies on PFAS and the introduction of
regulations. As of October 2023, only three PFAS are subject to
regulatory restrictions in Japan (PFOS, PFOA, and PFHxS), but
more are subject to restrictions in the USA and Europe.
Throughout the world, an increasing number of PFAS are
expected to require analysis. These changes will create more
analytical data that will have to be processed, creating a need for
more efficient methods of data analysis.
Peakintelligence is a peak integration software equipped with
peak-finding algorithms that have been developed with the help
of artificial intelligence (AI). This Application News describes a
case example in which it was used to process chromatogram data
of PFAS analysis and reduce the effort required and increase the
efficiency of the analysis.
• Peakintelligence is a product developed through joint research between
Shimadzu and Fujitsu.
Limitations of Previous Algorithms
Previous peak integration algorithms required the user to adjust
and fine-tune numerous detection and integration parameters
based on the chromatogram (Fig. 1). When the algorithms failed
to detect and integrate peaks correctly, manual corrections were
needed, which increased the workload for users. The skill level
required to carry out this work and the challenge of standardizing
this process also posed major issues when applying these
algorithms.
Fig. 1 Parameter Setting Window for Previous Peak Integration Algorithm
Peakintelligence Peak Integration Algorithms
Peakintelligence is a new peak integration technology that has
been developed using deep learning, a type of AI. AI is the name
used for the general concept and technology, while machine
learning and deep learning are the techniques on which AI is
built. Machine learning trains models with target attributes that
must first be extracted by a person, but deep learning trains
models with target attributes that are extracted mechanically
by software on a computer. For this reason, deep learning can
train models with large volumes of data and without the
variability caused by person involvement.
The trained model used by Peakintelligence was created from
approximately 13,000 chromatograms and with peak detection
and integration verified by experienced specialists. A training
dataset was prepared by labeling the chromatogram data and
peak start and end points, tuning the training and
hyperparameters, and assessing model performance (Fig. 2).
Installing this trained model on a laboratory PC gives the user
the ability to analyze LC-MS data automatically with minimal
effort.
• Peakintelligence does not have the functionality to perform additional
learning using newly acquired data.
Machine (PC)
Training
Trained model
Peakintelligence
Training
Output
Analysis
Trained PC
(Peakintelligence installed)
Fig. 2 Illustrative Overview of Peakintelligence
Input
Estimation