CheML
Chemometric Artifical Intelligence
Physics-Informed Data Amplification and Machine Learning Chemical Detection
Many standoff, mobile, and screening applications of chemical analysis cannot afford the benefit of physical or chemical separation steps to filter out potential interferences. CheML addresses the need to rapidly identify and quantity trace chemicals of interest in dynamic environments in the presence of interfering compounds.
It employs Artificial Intelligence to discern subtle pattern variations in chemical detector output. With CheML we synthesize tens of thousands of realistic laboratory chromatograms or spectra from small, physically acquired basis sets. CheML supports the virtual introduction of unlimited potential interferences and detector imperfections, enabling robust training of machine learning models that provide superior identification and quantification of trace compounds of interest.
AI developed within the CheML framework produces accurate quantitation in the presence of interferences, without the high level of human interaction and subjectivity required by conventional methods of analysis.
Applicable to numerous chromatographic and spectrometric methods
Saves enormous cost and time generating sufficient data for robust ML development
Produces robust immunity to unknown interferences
Results in fast, accurate, consistent data interpretation
Eliminates subjectivity in untangling signature overlap
Example synthetic chromatogram with target (green) and interferent (blue) compounds labeled T1-T5
and I1-I4, respectively.
Aggregated error between actual and CNN-predicted concentrations of five target compounds in the presence of consistent (left plot) or random (right plot) interferences. CNNs trained with interferences were more accurate than CNNs trained without.