Chromatography UV Peak Detection

Our Special Projects Intern, Eliott Park, presented his Chromatography UV Peak Detection model using a Keras TensorFlow neural net API. Promising results, despite the lack of primary data to train/test the model. With a little more time and data, this may be a viable application of machine learning in biotech manufacturing.

I’m amazed at the quality of work Eliott delivered in such a short internship period. It was a highly collaborative effort. He clarified our misconceptions of the tech as we evaluated possible real-world applications. Chrom UV peak detection was selected as we’re very familiar with the process and programming challenges.

The largest challenge getting sufficient data. Initially, we simulated elution data with various randomizing techniques. Eliott developed a viable work around using statistical methods to increase the number of realistically distributed training samples.

Lack of data highlights a broader issue with training biotech process models. A lot of data is required, but actual batch runs are expensive.

If you have access to a large, anonymized, set of UV values during an elution please send a message. It would be interesting to see how the model performs with actual data.