Rapid Copolymer Analysis of Unresolved Mass Spectra by Artificial Intelligence

In this Article, we present a novel data analysis method for the determination of copolymer composition from low-resolution mass spectra, such as those recorded in the linear mode of time-of-flight (TOF) mass analyzers. Our approach significantly extends the accessible molecular weight range, enabling reliable copolymer composition analysis even in the higher mass regions. At low resolution, the overlapping mass peaks in the higher mass range hinder a comprehensive characterization of the copolymers. Our approach, however, extracts the hidden information on the unique macromolecular chains embedded in these unresolved peaks. Regression relationships were developed between easily computable mass spectral parameters derived from the measured data and key characteristics of the copolymer structure including the average mole fraction, the number-average repeat unit count, and the monomer distribution. Two regression models were constructed: one using conventional statistical methods and the other employing a machine learning approach based on an Artificial Neural Network (ANN). The spectrum evaluation process is demonstrated for the analysis of various poly(N-acryloylmorpholine)-block-poly(N-isopropylacrylamide) (PNAM-b-PNIPAM) diblock copolymers. Our simplified approach is validated using experimental data and by comparison with our recently developed and reported highly computationally demanding method.