Pattern Recognition Studies of Complex Chromatographic Data Sets

Area of Science:

• Analytical Chemistry
• Chemometrics
• Bioinformatics

Background:

• Chromatographic fingerprinting of biological samples is complex.
• Multivariate statistics and pattern recognition aid analysis.
• Confounding variables and chance classification complicate sample differentiation.

Purpose of the Study:

• To present methods for analyzing complex chromatographic data.
• To address confounding experimental variables and chance classification.
• To demonstrate pattern recognition in biological sample classification.

Main Methods:

• Utilized ADAPT software for clustering, mapping, and statistical analysis.
• Applied pattern recognition to pyrochromatograms (PyGCs) of skin fibroblasts.
• Employed logistic regression and Monte Carlo simulations for data analysis.

Main Results:

• Successfully differentiated cystic fibrosis carriers from normal donors using PyGCs.
• Analyzed fire ant hydrocarbon profiles to relate to social caste and colony.
• Monte Carlo simulations established limits for chance classification probabilities.

Conclusions:

• Pattern recognition effectively classifies complex biological samples despite confounding factors.
• Developed discriminants for accurate sample classification in real-world studies.
• Established robust analytical approaches for chromatographic data analysis.