Chromatography plays a crucial role throughout industry and society. In recent decades, the technique has seen tremendous advancements. However, the technological progress remains unused to date. Exploiting the full technological capabilities of this state-of-the art separation technology requires deep knowledge and great time investments.
Our mission is to improve the accessibility of state-of-the-art separation technology. It is our vision that the computer algorithms under development will facilitate innovation in research and society.
We develop method optimization strategies that aim to simultaneously optimize the large number of parameters involved in chromatographic methods. Our approach relies based on direct communication between our algorithms and the analytical instrument. Our optimization strategy in development employs automated data analysis, retention modelling and Bayesian optimization, the latter which is a machine learning technique.
The AutoLC platform represents our prototype automated method development technology. The first iteration was published in 2022 in Analytical Chemistry. An important premise is our conviction that our algorithms must be flexible and modular so as to be able to hyphenate with algorithms developed by other scientists in the world.
The key characteristics of the AutoLC framework are:
Our labs at the Van ‘t Hoff Institute for Molecular Sciences of the University of Amsterdam feature several state-of-the-art liquid chromatography systems that are directly controlled by the AutoLC algorithm.
Our automated method development facility is globally unique and presents an exciting research environment. The systems can be used to test the latest developments and applications. The units feature dedicated custom-made detectors and separation technology to contribute to fundamental advances in chromatography, while also offering a paradise for programmers interested in progressing the application of chemometrics and statistics to this work.
Program for the interpretive optimization of two-dimensional resolution, B.W.J. Pirok, S. Pous-Torres, C. Ortiz-Bolsico, G. Vivó-Truyols and P.J. Schoenmakers, J. Chromatogr. A, 2016, 1450, 29–37, DOI: 10.1016/j.chroma.2016.04.061
In this paper, we explained why automated method development was the only viable route to improve the proliferation of advanced separation systems such as 2D liquid chromatography. We adopted the known concept of building retention models using data obtained from so-called scouting gradient methods and extended this to 2D separations. The relative simple study demonstrated a theoretical proof-of-principle supported by experiments. Important criticisms to this work included that the algorithm was not capable to run unsupervised, and that its retention models required peak tables that had to be manually created by the user.
Peak-tracking algorithm for use in comprehensive two-dimensional liquid chromatography – application to monoclonal antibody peptides, R.A. Molenaar, T.A. Dahlseid, G. Leme, D.R. Stoll, P.J. Schoenmakers, B.W.J. Pirok, J. Chromatogr. A, 2021, 1639, 461922, DOI: 10.1016/j.chroma.2021.461922.
One major criticism of the 2016 optimization workflow was the that the user was required to supply curated peak tables to the algorithm. In this study, an subroutine was developed to the AutoLC algorithm that could automatically track peaks across two comprehensive 2D-LC-MS chromatograms and thus generate the peak tables automatically.
Chemometric Strategies for Fully Automated Interpretive Method Development in Liquid Chromatography T.S. Bos, J. Boelrijk, S.R.A. Molenaar, B. van ‘t Veer, L.E. Niezen, D. van Herwerden, S. Samanipour, D.R. Stoll, P. Forré, B. Ensing, G.W. Somsen, B.W.J. Pirok, Anal. Chem. 2022, 94(46), 16060–16068, DOI: 10.1021/acs.analchem.2c03160
In this paper we proposed the AutoLC framework. We demonstrated for the first time a fully closed-loop optimization of an LC-MS separation of an antibody digest sample using retention modelling. We also showed that our modular algorithm could be used in combination with Bayesian optimization, which is a machine learning tool. This was the first prototype of the AutoLC algorithm, and developed as part of the UPSTAIRS project.
Closed-loop automatic gradient design for liquid chromatography using Bayesian optimization, Boelrijk, B. Ensing, P. Forré, B.W.J. Pirok, Anal. Chim. Acta, 2023, 1242, 340789, DOI: 10.1016/j.aca.2023.340789
In this work, we developed the Bayesian optimization variant of AutoLC. For this work, Jim Boelrijk developed a special kernel for the algorithm together with Dr. Patrick Forré.
Algorithm for tracking peaks amongst numerous datasets in comprehensive two-dimensional chromatography to enhance data analysis and interpretation, S.R.A. Molenaar, J.H.M. Mommers, D.R. Stoll, S. Ngxangxa, A.J. de Villiers, P.J. Schoenmakers, B.W.J. Pirok, Chromatogr. A, 1705, 2023, 464223: DOI: 10.1016/j.chroma.2023.464223
This study encompassed the development of an algorithm to conduct peak tracking for larger number of datasets. The resulting subroutines are employed by the AutoLC algorithm to process (and revisit) larger numbers of dataset at the same time. The work was conducted in collaboration with the University of Stellenbosch (Prof. André de Villiers), Gustavus Adolphus College (Prof. Dwight Stoll), and John Mommers (Envalior).
Computer-driven optimization of complex gradients in comprehensive two-dimensional liquid chromatography, S.R.A. Molenaar, T.S. Bos, J. Boelrijk, T.A. Dahlseid, D.R. Stoll, B.W.J. Pirok, Chromatogr. A, 2023, 1707, 464306, DOI: 10.1016/j.chroma.2023.464306
In this work we extended the AutoLC retention modelling variant (AutoLC-RM) to comprehensive 2D-LC separations. The work was applied on a RPLC×RPLC separation of complex antibody digest sample. This project was conducted by Stef Molenaar under the supervision of Bob Pirok in collaboration with the group of Prof. Dwight Stoll of Adolphus College.
We believe that our success is rooted in the interdisciplinary efforts between the worlds of analytical chemistry and computer sciences. We try to be as comprehensive as possible in our approach. From algorithms to automatically process data, fundamental retention modelling studies and dedicated detectors to track the shape of the programmed LC gradient.
We are grateful for our interactions within the Faculty of Science between the Informatics Institute and the Van ‘t Hoff Institute for Molecular Sciences, as well as our interactions with industry and other international groups such.
Bob Pirok of the Van ‘t Hoff Institute for Molecular Sciences at the University of Amsterdam has won the 2024 HTC Innovation Award, in recognition of his cutting-edge approaches to automate method development in analytical chemistry using machine learning and chemometrics. He received the award last Friday at the Hyphenated Techniques in Chromatography (HTC-18) conference in Leuven, Belgium.
A team of CAST scientists extended the AutoLC framework to 2D-LC-MS separations and applied it to the unsupervised optimisation of a 2D-LC separation of a peptide-digests.
Scientist of the Pirok group within CAST have managed to demonstrate a fully unsupervised method-development workflow for LC-MS. The algorithm was capable to optimize an antibody digest separation.
In the interpretive AutoLC framework, the algorithm establishes an initial understanding of the parameter space using pre-defined and generalized scouting methods (PHASE I). Then for each subsequent experiment, the algorithm preprocesses the data (PHASE II), interprets the chromatographic signals (PHASE III), constructs and updates either retention or machine learning models (PHASE IV) to then optimize the method further (PHASE V). As long as the objective criteria is not met, the new optimal method is automatically programmed into the LC by the algorithm and the loop continues.
The 2022 publication demonstrating the AutoLC workflow for the first time was dedicated to Prof. Peter Schoenmakers on the occasion of his retirement from the University of Amsterdam. Peter started his academic career with a pioneering paper on interpretive optimization in RPLC. Using a simple Texas Instruments calculator, he developed what can be seen as a very early prototype of what we present in this manuscript today.
Back in 1978, Peter envisaged the use of computer science to exploit chromatographic theory. We feel that this work essentially reflects Peter’s visionary conviction and are grateful for the inspiration he continues to provide.
This work, but also the great works of Kirkman & Snyder and many others, as well as commercial applications that resulted from this have been an inspiration to the AutoLC project.
