Richard Allmendinger

Efficient discovery of anti-inflammatory small-molecule combinations using evolutionary computing

The control of biochemical fluxes is distributed and to perturb complex intracellular networks effectively it is often necessary to modulate several steps simultaneously. However, the number of possible permutations leads to a combinatorial explosion in the number of experiments that would have to be performed in a complete analysis. We used a multi-objective evolutionary algorithm (EA) to optimize reagent combinations from a dynamic chemical library of 33 compounds with established or predicted targets in the regulatory network controlling IL-1β expression. The EA converged on excellent solutions within 11 generations during which we studied just 550 combinations out of the potential search space of ~ 9 billion. The top five reagents with the greatest contribution to combinatorial effects throughout the EA were then optimized pairwise. A p38 MAPK inhibitor with either an inhibitor of IκB kinase or a chelator of poorly liganded iron yielded synergistic inhibition of macrophage IL-1β expression. Evolutionary searches provide a powerful and general approach to the discovery of novel combinations of pharmacological agents with potentially greater therapeutic indices than those of single drugs.

Reference Point-Based Particle Swarm Optimization Using a Steady-State Approach

Conventional multi-objective Particle Swarm Optimization (PSO) algorithms aim to find a representative set of Pareto-optimal solutions from which the user may choose preferred solutions. For this purpose, most multi-objective PSO algorithms employ computationally expensive comparison procedures such as non-dominated sorting. We propose a PSO algorithm, Reference point-based PSO using a Steady-State approach (RPSO-SS), that finds a preferred set of solutions near user-provided reference points, instead of the entire set of Pareto-optimal solutions. RPSO-SS uses simple replacement strategies within a steady-state environment. The efficacy of RPSO-SS in finding desired regions of solutions is illustrated using some well-known two and three-objective test problems.

Closed-loop optimization of chromatography column sizing strategies in biopharmaceutical manufacture.

BACKGROUND This paper considers a real-world optimization problem involving the identification of cost-effective equipment sizing strategies for the sequence of chromatography steps employed to purify biopharmaceuticals. Tackling this problem requires solving a combinatorial optimization problem subject to multiple constraints, uncertain parameters, and time-consuming fitness evaluations. RESULTS An industrially-relevant case study is used to illustrate that evolutionary algorithms can identify chromatography sizing strategies with significant improvements in performance criteria related to process cost, time and product waste over the base case. The results demonstrate also that evolutionary algorithms perform best when infeasible solutions are repaired intelligently, the population size is set appropriately, and elitism is combined with a low number of Monte Carlo trials (needed to account for uncertainty). Adopting this setup turns out to be more important for scenarios where less time is available for the purification process. Finally, a data-visualization tool is employed to illustrate how user preferences can be accounted for when it comes to selecting a sizing strategy to be implemented in a real industrial setting. CONCLUSION This work demonstrates that closed-loop evolutionary optimization, when tuned properly and combined with a detailed manufacturing cost model, acts as a powerful decisional tool for the identification of cost-effective purification strategies.

On handling ephemeral resource constraints in evolutionary search

We consider optimization problems where the set of solutions available for evaluation at any given time t during optimization is some subset of the feasible space. This model is appropriate to describe many closed-loop optimization settings (i.e., where physical processes or experiments are used to evaluate solutions) where, due to resource limitations, it may be impossible to evaluate particular solutions at particular times (despite the solutions being part of the feasible space). We call the constraints determining which solutions are non-evaluable ephemeral resource constraints (ERCs). In this paper, we investigate two specific types of ERC: one encodes periodic resource availabilities, the other models commitment constraints that make the evaluable part of the space a function of earlier evaluations conducted. In an experimental study, both types of constraint are seen to impact the performance of an evolutionary algorithm significantly. To deal with the effects of the ERCs, we propose and test five different constraint-handling policies (adapted from those used to handle standard constraints), using a number of different test functions including a fitness landscape from a real closed-loop problem. We show that knowing information about the type of resource constraint in advance may be sufficient to select an effective policy for dealing with it, even when advance knowledge of the fitness landscape is limited.

Multiobjective optimization: When objectives exhibit non-uniform latencies

Building on recent work by the authors, we consider the problem of performing multiobjective optimization when the objective functions of a problem have differing evaluation times (or latencies). This has general relevance to applications since objective functions do vary greatly in their latency, and there is no reason to expect equal latencies for the objectives in a single problem. To deal with this issue, we provide a general problem definition and suitable notation for describing algorithm schemes that can use different evaluation budgets for each objective. We propose three schemes for the bi-objective version of the problem, including methods that interleave the evaluations of different objectives. All of these can be instantiated with existing multiobjective evolutionary algorithms (MOEAs). In an empirical study we use an indicator-based evolutionary algorithm (IBEA) as the MOEA platform to study performance on several benchmark test functions. Our findings generally show that the default approach of going at the rate of the slow objective is not competitive with our more advanced ones (interleaving evaluations) for most scenarios.

Efficient discovery of chromatography equipment sizing strategies for antibody purification processes using evolutionary computing

This paper considers a real-world optimization problem involving the discovery of cost-effective equipment sizing strategies for the chromatography technique employed to purify biopharmaceuticals. Tackling this problem requires solving a combinatorial optimization problem subject to multiple constraints, uncertain parameters (and thus noise), and time-consuming fitness evaluations. After introducing this problem, an industrially-relevant case study is used to demonstrate that evolutionary algorithms perform best when infeasible solutions are repaired intelligently, the population size is set appropriately, and elitism is combined with a low number of Monte Carlo trials (needed to account for uncertainty). Adopting this setup turns out to be more important for scenarios where less time is available for the purification process.

On-line purchasing strategies for an evolutionary algorithm performing resource-constrained optimization

We consider an optimization scenario in which resources are required in order to realize or evaluate candidate solutions. The particular resources required are a function of the solution vectors, and moreover, resources are costly, can be stored only in limited supply, and have a shelf life. Since it is not convenient or realistic to arrange for all resources to be available at all times, resources must be purchased on-line in conjunction with the working of the optimizer, here an evolutionary algorithm (EA). We devise three resource-purchasing strategies (for use in an elitist generational EA), and deploy and test them over a number of resource-constraint settings. We find that a just-in-time method is generally effective, but a sliding-window approach is better in the presence of a small budget and little storage space.

Evolutionary optimization on problems subject to changes of variables

Motivated by an experimental problem involving the identification of effective drug combinations drawn from a non-static drug library, this paper examines evolutionary algorithm strategies for dealing with changes of variables. We consider four standard techniques from dynamic optimization, and propose one new technique. The results show that only little additional diversity needs to be introduced into the population when changing a small number of variables, while changing many variables or optimizing a rugged landscape requires often a restart of the optimization process.

Life‐cycle and cost of goods assessment of fed‐batch and perfusion‐based manufacturing processes for mAbs

Life‐cycle assessment (LCA) is an environmental assessment tool that quantifies the environmental impact associated with a product or a process (e.g., water consumption, energy requirements, and solid waste generation). While LCA is a standard approach in many commercial industries, its application has not been exploited widely in the bioprocessing sector. To contribute toward the design of more cost‐efficient, robust and environmentally‐friendly manufacturing process for monoclonal antibodies (mAbs), a framework consisting of an LCA and economic analysis combined with a sensitivity analysis of manufacturing process parameters and a production scale‐up study is presented. The efficiency of the framework is demonstrated using a comparative study of the two most commonly used upstream configurations for mAb manufacture, namely fed‐batch (FB) and perfusion‐based processes. Results obtained by the framework are presented using a range of visualization tools, and indicate that a standard perfusion process (with a pooling duration of 4 days) has similar cost of goods than a FB process but a larger environmental footprint because it consumed 35% more water, demanded 17% more energy, and emitted 17% more CO2 than the FB process. Water consumption was the most important impact category, especially when scaling‐up the processes, as energy was required to produce process water and water‐for‐injection, while CO2 was emitted from energy generation. The sensitivity analysis revealed that the perfusion process can be made more environmentally‐friendly than the FB process if the pooling duration is extended to 8 days.

Policy learning in resource-constrained optimization

We consider an optimization scenario in which resources are required in the evaluation process of candidate solutions. The challenge we are focussing on is that certain resources have to be committed to for some period of time whenever they are used by an optimizer. This has the effect that certain solutions may be temporarily non-evaluable during the optimization. Previous analysis revealed that evolutionary algorithms (EAs) can be effective against this resourcing issue when augmented with static strategies for dealing with non-evaluable solutions, such as repairing, waiting, or penalty methods. Moreover, it is possible to select a suitable strategy for resource-constrained problems offline if the resourcing issue is known in advance. In this paper we demonstrate that an EA that uses a reinforcement learning (RL) agent, here Sarsa(λ), to learn offline when to switch between static strategies, can be more effective than any of the static strategies themselves. We also show that learning the same task as the RL agent but online using an adaptive strategy selection method, here D-MAB, is not as effective; nevertheless, online learning is an alternative to static strategies.