By Connor Taylor on August 15, 2020
In the latest Compunetics build, now available for download, further functionality has been added to the software to guide experimentation for process optimisation. The new Compunetics Investigator is the first application of its kind that works in tandem with the user to optimise a chemical process, by requesting iterative experiments for the scientist to complete, to effortlessly optimise a bio/chemical system - without the need for any scientific modelling or expertise. Compunetics Investigator utilises a simplex optimisation algorithm, to quickly and efficiently identify optimum operating conditions.
Why use an algorithm?
Often when optimising a chemical process, researchers will conduct a 'one-factor-at-a-time' (OFAT) optimisation approach. This OFAT approach is a common method of optimisation, especially in academia, in which experiments are performed that are guided by scientific intuition, by fixing all process factors except for one. These factors are experimental conditions (such as temperature, reagent stoichiometry, reaction time etc.) which when combined, make a multi-dimensional space called the parameter space, containing a large number of possible combinations of these factors, each of which would be one experiment. This parameter space is constrained by the lower and upper limits of each factor (e.g. max and min temperature). After the best value for one factor has been optimised, another set of experiments are executed to then optimise another factor, until all factors are optimised and the scientist believes that they have arrived at the optimal reaction conditions. However, this method gives an incomplete picture of the chemical process, as it disregards any synergistic effects between any factors in the complex parameter space. This typically leads to conducting many more experiments than necessary, as well as inaccuracies in determining the optimum reaction conditions, which become even more inaccurate as the number of optimised factors increases.[2,3]
Using a machine learning algorithm (such as a simplex algorithm), on the other hand, is one of the most efficient techniques to obtain accurate and truly optimised reaction conditions. The super-modified simplex (simplex) algorithm starts by performing experiments that outline convex polyhedra of n+1 vertices, where n is the number of variables - for example, for a 2-variable optimisation, shapes with 3 vertices will be outlined (triangles). This polyhedron, or simplex, explores the feasible parameter space defined by the user, based on a defined response (yield, selectivity etc.). The worst performing vertex is then replaced by a specific geometric transformation, as the algorithm locates areas with a better response based on mathematical modelling, until a better response cannot be identified and the optimisation terminates. Further details on the specific algorithm components and transformations can be found in the original method publication. This type of machine-learning-guided optimisation is well documented in the chemical literature and is well known for consistently outperforming conventional optimisation techniques such as OFAT.
A representative comparison of OFAT vs. simplex optimisation for a chemical response contour is shown below, where numbers represent the experiment number. In this OFAT example, the temperature is first held constant whilst the reagent equivalents are optimised, and after experiment 5 is found to be optimal, the temperature is then optimised at a fixed reagent equivalent, arriving at the 'optimum' response at experiment 13. As this optimisation strategy does not incorporate interference-effects between factors, the true optimum is rarely found. In this simplex example, however, both factors are optimised in tandem as the machine learning approach determines the best experiments to conduct, hence identifying the true optimised reaction conditions.
Although this optimisation technique is reported in the chemical literature, primarily in the circles of automated optimisation (self-optimisation), Compunetics is the first and only accessible scientist-facing software for guided experimentation and optimisation. Individual reactions are suggested to the scientist, then adaptive modelling determines the most beneficial experiment to conduct in order to move towards and ultimately identify optimum reaction conditions. No machine learning or computer science knowledge required. No coding or modelling expertise necessary.
How do I use it?
Simply input the experimental variables of interest (and their corresponding lower and upper bounds) and whether you would like to maximise or minimise your response of interest. These variables could be anything, and the response can also be anything - as long as it can be measured. Examples could be: yield of product, suppression of side-product, enantiomeric excess, E-factor etc.
It is then simply a matter of conducting the required experiments that the machine learning algorithm suggests and recording the response. The dynamic algorithm will then use this information to deduce the next experiment for the scientist to conduct, therefore iteratively moving towards an optimised response. Then of course, after the optimisation has begun, an optimisation file will be saved and constantly updated so that you can continue the optimisation at a later date if necessary. Full details and examples can be found in the user guide with the Compunetics download.
Where can I get it?
The new Compunetics build is licensed in partnership with the University of Leeds and can be accessed via the Compunetics website. Compunetics is currently in beta testing and is free for all academic and commercial use. Windows PC required. MATLAB Runtime 9.7 specifically required.
How do I update my current Compunetics build?
To update, simply download the new Compunetics folder via the Compunetics website. The old Compunetics folder can simply be deleted and your new Compunetics.exe will run without the need for activation.
Additional updates in the current Compunetics build
1: Aggarwal, V.K., A.C. Staubitz, and M.R. Owen, Optimization of the mizoroki-heck reaction using design of experiment (DoE). Organic Process Research & Development, 2006. 10(1): p. 64-69.
2: Owen, M.R., et al., Efficiency by design: optimisation in process research. Organic Process Research & Development, 2001. 5(3): p. 308-323.
3: Wahid, Z. and N. Nadir, Improvement of one factor at a time through design of experiments. World Applied Sciences Journal, 2013. 21: p. 56-61.
4: Routh, M. Wm, P. A. Swartz, and M. B. Denton. Performance of the super modified simplex. Analytical Chemistry, 1977. 49(9): 1422-1428.
5: Clayton, A.D., Manson, J.A., Taylor, C.J., et al., Algorithms for the self-optimisation of chemical reactions. Reaction Chemistry & Engineering, 2019. 4(9), pp.1545-1554.
6: Algorithm developed by Dr Adam D. Clayton.