MSACL 2019 US Abstract

Automating HPLC Peak Detection using Convolutional Neural Networks

Matthew Boutell (1), Randall Julian (2)
(1) Rose-Hulman Institute of Technology, Terre Haute, IN (2) Indigo BioAutomation, Carmel, IN

Warning: Undefined variable $headshot in /var/www/html/view_abstract/view_abstract_in_program.php on line 704

Matthew Boutell (Presenter) Rose-Hulman Institute of Technology

Presenter Bio: Matt Boutell is Professor of Computer Science and Software Engineering at Rose-Hulman Institute of Technology in Terre Haute, Indiana. His primary area of research is machine learning applied to image recognition problems. He earned his PhD in computer science at the University of Rochester while researching machine learning techniques to improve organization of large digital photo collections for the Eastman Kodak Company. Matt was a Fulbright Scholar to Copperbelt University in Kitwe, Zambia in 2011-2012 and received the Dean’s Outstanding Teacher award from Rose-Hulman in 2016. His other professional interests include algorithms, Android app development, online education, and the effect of individualized education on learning.

Relevant Financial Disclosures (within past 24 months) Salary Indigo BioAutomation Discussing Products? NO, presenter will NOT present educational content related to products/services of the ineligible company(ies) listed above. Conflict mitigation NOT REQUIRED.

Visually detecting peaks in LC-MS chromatograms is a straightforward task for humans, but biases and pattern recognition skill variability cause problems with reproducibility. Algorithmic review of clinical raw data without human intervention solves this but is difficult. Existing algorithms model and fit peaks within a semi-automated workflow. To increase confidence in the results, we propose an independent algorithm that uses raw chromatograms as input and classifies peaks using a convolutional neural network, similar to those used for image-based diagnostics. It classifies chromatograms as either peak, no-peak, high-intensity-no-peak, or small-peak, obtaining accuracies of 88.8% overall, 97.1% on non-peaks, and 93.8% on peaks, on 2.3 million chromatograms.