Biologics refers to the use of biotechnology in therapeutic discovery and medicine. A biological drug, or a biologics, is a product created from a living organism produced by the body such as proteins and cell lines. Biologic medicine consists of large molecules and can be used to treat diseases such as cancer, diabetes, Crohn’s, cystic fibrosis, growth deficiencies, and many more. Such treatments include vaccines, antibodies, and hormones. Biologics are not new. In fact, the development of biologics has been around for about 20 years. But the way we can now manufacture them is new. And it was during Oxford Global’s Biologics EU 2021 event, held in April, that talks surrounding the possibility of their computational manufacturing started to heat up.
The Technology: How Computational Software Aids Biologics Discovery
Computational engineering is a relatively novel discipline in the world of biotherapeutic manufacturing and refers to the development and optimisation of therapeutics using computer-based software. In biologic drug production, computational techniques can be used to assist in the discovery of these therapies by imitating the pre-established design process and synthesis of these proteins. Computation engineering is divided into two main strands: Artificial Intelligence (AI) and Machine Learning (ML). For bioloigcs to be deemed successful, they must feature certain physiochemical activity and properties that can be globally recognised as possessing the potential for developability. Developability depends on whether these prerequisites can be simultaneously optimised to accommodate a broad design space of protein sequences and buffer compositions. AI and ML can accelerate and improve this process. By increasing the activity, effectiveness, and safety of the protein properties, computational techniques like these can in turn significantly reduce their manufacturing time and costs.
The Dark Art of Biologics: The Evolution of Scientific Computer Programming
With the advisory board deeming computational engineering a ‘dark art’ of biologics development, we were intrigued to find out more about this rapidly expanding discipline. Caroline Barelle, founder and CEO of the antibody and next-generation therapeutics company, Elasmogen, explained how “there are definitely a lot of things happening in the [biologics] world with artificial intelligence, augmented intelligence, and design”. She continued, “it is an area that is going to expand” with the reliance on scientific computer programming “really taking off and com[ing] of age”.
For Shane Olwill, Senior Vice President of Translational Science at Pieris, digitisation and AI in the biologics world presents an exciting array of opportunities. According to Olwill, computational science is “something that cuts across multiple aspects and disciplines”. He claimed how “these types of tools are today being applied to other aspects of biotechnology and not just the design of your target, target antibodies, your target proteins, but design and management of manufacturing processes, for example”.
Machine Learning in Protein Design: A Case Study
One such scientist to adopt a computer-based approach to research is Professor Phillip M. Kim of the University of Toronto. Kim leads a research laboratory at the university’s Donnelly Centre and integrates machine learning with physics-based modelling methods for the engineering of biologics. Speaking at the Biologics EU event, he discussed the benefits of using ML in protein design. By using a graph neural network-based model, Kim and his team of scientists were able to teach the neural network what a protein looks like and then supply the neural network a new fold in its matrix to make a real novel protein. “The graph neural network-based model generates novel proteins that adopt the pre-determined shape with remarkable accuracy”, Kim stated.
He also highlighted how integrated computational designs allow for specificity during the selection process. “There are certain advantages because computational design ensures targeting of particular interface/conformation – we are not just subject to whatever our selection gives us, we can target specific conformational sites”, Kim contends. “Computational designs enable full control over the sequence, can directly build in favourable properties for developability – so we can build into it all the properties we want to”, he continues.
A Frontier Too Far?: The Difficulties of Computational Engineering
Whilst for Kim, the encouraging results of early ML engineering “help[s] pave the way for better developability later on”, there is talk about its possible limitations. The complexity of computational design is at the forefront of discussion. Associate Director of AstraZeneca, Rick Davies, points out that the relative infancy of integrated computational engineering may become a bottleneck to its systematic advancement. He states, “the thing that we’re lacking in order to make the model better is data in a sufficiently formatted way that we can actually use for machine learning”.
Another challenge faced by scientists working within bioengineering is the ease with which structural biology can be applied to technological processing. Using AI algorithms may compromise the desired pre-determined shape of the biologic product. In other words, we must ask whether the product of the computationally designed biologics is in fact correct. It is imperative that these AI algorithms and programming are accurate and can provide the right structure.
The future role and advancements in computational engineering of biologic therapies remain highly progressive. With an increasing number of pharmaceutical companies and university laboratories delving further into the field of AI and ML we expect to see some exciting research and collaborations at Oxford Global Conferences. This time next year will machines be running the show within biologics drug discovery? Only time, and technology, will tell.