The rush to digitalization within Biopharma has resulted in the development and availability of new capabilities and technologies, both internally and through external service providers. This can only benefit the industry but creates a new challenge – having set your digitalization plan and the objectives, which approach or service to use? Ever heard of the no-free-lunch theorem? In a nutshell, this is the machine-learning version of “it depends”. There is no one-size-fits-all approach.
A crucial first step to decide the appropriate approach is to assess the requirements to reach your plans objective. For instance, if the dynamic evolution of a process should be assessed (the requirement), then static models will be of limited help (exclusion criteria). Additionally, the listing of the requirements will create perspective on the set objective – were they too narrowly defined perhaps? These early observations payback tenfold, allowing for corrective adjustments before resources are committed.
With clarity on requirements, options can be assessed from a conceptual point of view. Using the dynamic evolution as an example, possible approaches include: a linear state space model, dynamic machine-learning models, mechanistic models, or hybrid models, besides others.
Selection of the most appropriate model are guided by the following questions:
1) What is the expected effort to develop the approach?
2) Is it likely that the selected approach is successful?
3) What information is available?
Create a benchmark
From a practical point of view, one could start by assessing the simplest approach (given the information available) and generate a benchmark upon which alternatives are assessed. For instance, if a mechanistic model of the process was available (e.g. from literature) yet relatively little amount of process data, the creation of a linear state space, machine-learning, or hybrid model would likely require more effort than using the mechanistic model first.
Once the results from the benchmark approach are available, a reflection on the quality of results should be carried out, also considering additional efforts that might be needed to improve the results. Are the results good enough?
Call in the experts when needed
Some approaches, be the fact they are a new innovation, might be unfamiliar and which you may not fully understand. In such case, try to understand why this approach might be better. In some cases, the most efficient way to do this might be for asking for external advice from the innovation experts. Dismiss upstart innovation at your own risk – what is new today could be market standard tomorrow.
Innovation leads, and project managers have more options then ever when it comes to tools and solutions to fill their digitalization agenda. While tempting to just dive into the innovation pool, spending appropriate time assessing the options based on their ability to meet your plan's objective will pay back dividends for years thereafter.
Write a comment