Achieving optimal SERS through enhanced experimental design

Abstract

One of the current limitations surrounding surface-enhanced Raman scattering (SERS) is the perceived lack of reproducibility. SERS is indeed challenging, and for analyte detection, it is vital that the analyte interacts with the metal surface. However, as this is analyte dependent, there is not a single set of SERS conditions that are universal. This means that experimental optimisation for optimum SERS response is vital. Most researchers optimise one factor at a time, where a single parameter is altered first before going onto optimise the next. This is a very inefficient way of searching the experimental landscape. In this review, we explore the use of more powerful multivariate approaches to SERS experimental optimisation based on design of experiments and evolutionary computational methods. We particularly focus on colloidal-based SERS rather than thin film preparations as a result of their popularity. © 2015 The Authors. Journal of Raman Spectroscopy published by John Wiley & Sons, Ltd.

Keywords: SERS; chemometrics; design of experiment; genetic algorithm; optimisation.

Figure 1

Schematic depicting which variables should be considered when optimising a system for achieving the best surface‐enhanced Raman scattering response in terms of the signal enhancement being strong, robust and reproducible. This schematic highlights the individual parameters, various characterisation techniques as well as the data analysis approach that need to be considered during the design of experiments phase. FWHM, full width half maximum; PCA, principal component analysis; SD, standard deviation.

Figure 2

A 3D representation of a typical design of experiment that incorporates different parameters that needs to be optimised in order to achieve the most optimal surface‐enhanced Raman scattering (SERS) response. Sequential rounds of assessment are performed until a realistic solution(s) is obtained. The concept of Pareto optimality is demonstrated: when optimising parameters within a SERS experiment, it is not necessarily true that the conditions optimal to signal enhancement are also optimal to reproducibility and so a trade‐off between the two objectives must be established. As an example, the parameters to be optimised concurrently include parameter 1 = pH, parameter 2 = concentration of colloid and parameter 3 = aggregating agent.

Figure 3

An illustration of the genetic algorithm (GA) approach to surface‐enhanced Raman scattering (SERS) optimisation using a mountain analogy. A GA reduces the number of steps taken in order to reach the highest fitness value through the evolution of solutions to find the highest peak. (A) Denotes a simple fitness landscape (e.g. Mount Fuji), where this may mean that only a single variable needs optimising. In this case, a GA approach is not necessarily required, and simple hill climbing algorithms would suffice. Whereas (B) highlights a more complex solution where multiple routes can be taken to reach the summit of the fitness landscape, i.e. multiple variables need to be optimised simultaneously, until the highest point is reached (here depicted by the Himalayas), and the application of GA may simplify the number of solutions to reach the optimum fitness value. The figures are available from the Creative Commons license agreement. For more details, see Mount Fuji image on Flickr https://www.flickr.com/photos/9177053@N05/4469232631/in/photostream/ and Himalayas image on Deviant art http://citizenfresh.deviantart.com/art/Himalaya‐Mountains‐1‐Nepal‐72353246

Figure 4

A schematic of the overall evolutionary approach that is used in experimental design. (A) Outlines a workflow of the genetic algorithm (GA) approach. The procedure continues to evolve until the population converges or when a maximum number of iterations is reached. (B) Denotes the methods used to generate a new population. These include offspring (in this example tigons and ligers) from the parents (a lion and a tiger). Mutations can also be generated and introduced from parents; e.g. mutant 1 is generated from a single‐point mutation to parent 2. This GA process needs translation in terms of surface‐enhanced Raman scattering optimisation: in the example provided, child 2 is translated as a solution where a Au citrate‐reduced colloid with NaCl as the reducing agent is used for surface‐enhanced Raman scattering.