High-throughput (HT) experimentation is transforming biotechnology by enabling systematic exploration of complex multi-dimensional experimental conditions. However, current analytical methods are often unable to handle the rapid pace of sample generation in HT workflows. This study presents an integrated system of physical devices and software to automate and accelerate Raman spectral measurements in HT-facilities. The setup simultaneously handles eight parallel 50 μL samples delivered by a pipetting robot, completing measurement, handling, cleaning, and concentration prediction within 45 s per sample. We introduce a machine learning model to predict metabolite concentrations from Raman spectra, achieving mean absolute errors of 0.27 g L −1 for glucose and 0.06 g L −1 for acetate during Escherichia coli cultivations. This approach enables consistent high-throughput spectral data collection for fermentation monitoring, calibration, and offline analysis, supporting the generation of extensive datasets, enabling the training of more robust and generalizable machine learning models.
Lange et al. (Fri,) studied this question.