Combinations of Analytical and Machine Learning Methods in a Single Simulation Framework for Amphoteric Molecules Detection

The most recent advances in personalized medicine require highly accurate drug syntheses. A significant component of synthesis is verification. This step requires confirming the sequence of amino acids (AA) in the target drug (protein). This paper presents a novel methodology for identifying amphoteric molecules such as amino acids, peptides and many enantiomers using the signal from the surface potential and capacitance obtained from Field Effect Transistor (FET) sensors. Our methodology combines the Site-binding and Gouy-Chapman-Stern (GCS) models with a transformer mode. We have termed the resulting approach BioTokens. It can find AA sequences up to a length of 5 with 95% accuracy. BioTokens achieves this by using FET sensor data, which is cheaper and easier to obtain than the competing state-of-the-art mass spectroscopy technology. This paper presents the BioTokens concept and shows our initial steps in its development.