Synthetic Electrochemical Biosensor Data with Conditional Variational Autoencoders for Enhanced Predictive Modeling
Desmond Kai Xiang Teo, Tomás Maul, Michelle T.T. Tan · 2024
Electrochemical biosensors (EB) have revolutionized various domains with their rapid and sensitive analyte detection capabilities. However, as EBs advance to tackle more challenging targets, the complexity of post hoc analysis intensifies, posing a significant bottleneck. Integrating AI with biosensing devices offers a promising solution, yet widespread adoption faces hurdles such as limited data availability. This study introduces a novel approach leveraging Conditional Variational Autoencoders to generate synthetic electrochemical data for augmenting training datasets. This approach aims to address the specific issue of data scarcity in the post hoc analysis of electrochemical biosensing data. Evaluation experiments demonstrate the effectiveness of incorporating synthetic data in enhancing prediction model performance, reducing the MRE by 41.16%, and achieving similar performance to a model trained with the full dataset by using only 50% of the training data supplemented with synthetic data. This work highlights a critical advancement, bridging the gap between AI and electrochemical biosensing, ultimately improving the efficiency and accuracy of post hoc analysis in this vital field.