Automated GC/MS Characterization of Ball Point Pen Inks by Differential Analysis and Predictive Modeling

Significance of the Topic

Determining the age and origin of ball-point pen inks on paper is critical in forensic investigations, document authentication, and quality control. Trends in ink chemistry over time and subtle compositional changes make manual interpretation challenging. Integrating automated sampling with advanced chemometric modeling offers a reproducible, high-throughput route to ink classification and dating.

Objectives and Study Overview

This study aimed to develop and validate a fully automated GC/MS workflow, coupled with statistical class prediction, to differentiate ball-point pen inks and estimate the time since deposition. Key goals were:

• Rapid sample introduction and data acquisition using Thermal Separation Probe (TSP) and GC/MS.
• Automated spectral deconvolution and entity extraction.
• Application of multiple predictive algorithms for ink classification and age estimation.
• Assessment of model performance on unknown samples.

Methodology and Instrumentation

Sample Preparation

• Copier paper squares (1.5 cm) were scribbled with three different ball-point pen inks.
• Samples rolled into TSP microvials, spiked with 1 µL internal standard.
• Replicates prepared at various aging intervals and stored in inert containers.

Instrumentation

• Agilent 7890 GC with 5975C MS detector.
• Multi-mode inlet (MMI) with Thermal Separation Probe.
• Column: DB-5HT, 15 m, He at 1.2 mL/min.
• Temperature programs: Inlet 100 °C to 280 °C; Oven 70 °C to 340 °C.
• MS conditions: source 240 °C, quadrupole 150 °C, scan range 40–570 m/z.

Data Processing

• Automated spectral deconvolution and entity extraction using AMDIS.
• Differential analysis and filtering in Mass Profiler Professional (MPP): abundance threshold, RT tolerance, ANOVA (p < 0.05).
• Class prediction models built and tested via ChemStation Automation (SCP 2.0).

Main Results and Discussion

• Initial feature set of 341 entities was filtered to 87 ink-specific markers.
• Five statistical models were evaluated on five unknown samples:

• Support Vector Machine (SVM)
• Partial Least Squares Discrimination (PLS-DA)
• Neural Network
• Decision Tree
• Naïve Bayes

• PLS-DA and SVM correctly classified all unknowns, with confidence levels ranging from 53 % to 85 % (PLS-DA) and up to 67 % (SVM).
• Neural Network and Naïve Bayes each identified 2 of 5 unknowns; Decision Tree identified 4.
• Workflow automation reduced user intervention to sample loading and sequence start, enabling consistent, reproducible results across instruments.

Benefits and Practical Applications

• High throughput and minimal manual intervention accelerate forensic casework.
• Robust chemometric filtering improves reliability when analyzing complex ink profiles.
• Flexible model evaluation allows selection of optimal algorithm for specific ink types.
• Methodology transferable to quality assurance in manufacturing and document verification.

Future Trends and Potential Applications

• Expanding the training dataset with diverse ink brands and aging periods to enhance model confidence.
• Integration of high-resolution MS and alternative ionization sources to capture additional chemical markers.
• Development of cloud-based prediction platforms for real-time forensic decision support.
• Application to other forensic matrices, such as toner or textile dyes.

Conclusion

This work demonstrates that a fully automated GC/MS workflow, combined with rigorous chemometric filtering and predictive modeling, can reliably classify ball-point pen inks and estimate sample age. The approach minimizes operator involvement, ensures reproducibility, and offers a versatile foundation for forensic and industrial ink analysis.