Highlights from Recovery XVIII

Impression from Recovery XVIII conference site in Asheville, NC, USA

Various ChromX applications presented by Genentech, Roche, RPI, and GoSilico

GoSilico’s co-founders Thiemo Huuk and Tobias Hahn had the honor to be invited to their second and third “Recovery of Biological Products Conference”, respectively. Recovery XVIII took place in Asheville, NC, USA, October 7-12, 2018. The theme “Bioprocessing 3.0” brought together top academic and industrial scientists and engineers to discuss product design, process development, manufacturing, quality and data management, analytics, etc.

In the following, we want to share some personal highlights from GoSilico’s perspective, that is model-based chromatography process development and characterization. This is just a small excerpt and should not devalue the other contributions in any way.

Fundamentals of Chromatography Modeling

There is some progress in the research on adsorption isotherms. Prof. Lenhoff (U. Delaware) presented a salt-dependent version of his colloidal model for ion-exchange chromatography along with a calibration strategy based on breakthrough experiments. A major advantage is the ability to describe asymmetric elution peaks as also reported by GoSilico and Roche.

Prof. Przybycien (RPI) focused on pH-modulated Protein A affinity chromatography in his presentation. The initial results are promising and we look forward to an extension to include changes in ionic strength and, of course, to further support his research with our ChromX software.

For both cases, IEX under high-load conditions and Protein A, GoSilico has been successful in the past using in-house approaches. But it is always good to challenge the existing models and workflows.

GoSilico’s contributed to this area with a poster presentation on extended models for hydrophobic interaction chromatography that could account for pH changes a mixed-mode binding. Future applications of these models include e.g. pH-tuned HIC processes, as they are frequently applied in industry.

Application of Models in Process Development and Process Characterization

A first long-anticipated model-based linkage of step study was presented by Jessica Yang from Genentech in her talk “Process Clearance Predictions using Mathematical Modeling”. The experimental effort to perform a linkage study increases of course exponentially with the considered model parameters. It is practically infeasible to screen the whole process operating space such that usually only extreme cases are considered that do not really reflect typical conditions. With the help of mechanistic models, it is straight forward to connect two independently calibrated models and simulate thousand, ten-thousands or even more process scenarios. Needless to say, that Jessica used ChromX for her computations. This helps to generate process understanding, avoids over-optimization of earlier unit operations and could potentially lead to accelerated process validation in the QbD paradigm.

A lot of space in the Recovery program was also reserved for new modalities such as viral vectors and gene therapy. While the molecule candidates are just reaching clinical trials and some quality attributes might yet be unknown, all of the presented purification approaches included at least one chromatography step. As those new formats cannot just use Protein A capture as for monoclonal antibodies, process development might get more complicated – at least when using traditional DoE approaches. In the course of their PERMIDES project, CEVEC and GoSilico have shown that developing a capture step for a Virus-like Particle (VLP) can be done with minimal effort in silico. In an ion-exchange case study, only three experiments were needed to develop the capture step, two of which were historical runs at low load challenge. The talk by Tobias Hahn also showed that the widely used SMA model still holds for molecules of VLP size, and that characteristic charge values can reach 25 and above.

Uncertainty and Quality Management

In bioprocessing, uncertainty is unavoidable and need not be necessarily bad if it is identified, quantified and controlled. On a broader scope this was very visually explained in the excellent key note by Jeffrey Baker (FDA). With focus on mechanistic modeling, Ferdinand Stückler (Roche) showed that especially Bayesian statistics can be very valuable to understand the influence of uncertainty on process performance and quality attributes. Ferdinand also tackled the problem that during model building, different parameter sets might be equally well suited to describe data and it might not be obvious which of the “local minimums” might be the true parameter set. His solution based on ChromX and treated all of the candidates equally to generate a distribution of possible outcomes. If the outcomes are within the quality criteria, the remaining uncertainty is considered acceptable.

The talk by Tobias Hahn (GoSilico) presented a Quality-by-Design (QbD) guided approach to model building that aims at a related problem. The presented workflow aims to ensure model quality by design, especially keeping model parameters mechanistically sound. This way, we avoid running into several local minimums in the first place. It should also be mentioned, that we at GoSilico always check if all our model parameters are well identified and if they serve a unique purpose. If not, more laboratory experiments should be performed or model parameters eliminated. A very useful tool in this respect is the method of Optimal Experimental Design (OED) that is part of GoSilico’s chromatography simulation software ChromX.


A very stimulating talk was given by Prof. Cramer from RPI, with the goal to start a broad collaboration in academia and industry to make all the data accessible that is generated in the industry. Most often it cannot be re-used because of differences in data collection, data storage or simple naming convention. A typical example from our daily work at GoSilico is that we always have to clarify first, what different companies and vendors understand when talking about porosities and capacities. For example, a capacity in ChromX is a packing-independent material parameter that should be calculated with respect to the solid volume in a chromatography column and not the whole inner column volume. Such simple difference can hinder the application of all the great tools developed under the terms machine learning and artificial intelligence.

About the Recovery Conference series

From the website: “The Recovery of Biological Products Conference Series is the premier international forum for the presentation and discussion of recent advances in the operations used to recover biological products of therapeutic, diagnostic and industrial value to society. An invitation-only biennial event, the Conference Series provides delegates with a clear and comprehensive understanding of the status and direction of this fast-moving field.”

Expression of Thanks

Last but not least, Thiemo and Tobias want to thank the conference and session chairs for the invitation to this perfectly organized and truly inspiring event. We are also very grateful for the contributions by CEVEC, CSL, GE, KIT and Roche that stimulated and improved our presentations substantially. Thanks also to Genentech, Roche and RPI for showcasing their ChromX results.

Do NOT follow this link or you will be banned from the site!