Over the years, disease diagnosis, drug development and drug testing have become increasingly inefficient and expensive. Finding a successful way to reduce the cost and time spent by pharmaceutical companies during the drug discovery and clinical trial phases of drug development is paramount to ensuring efficacious and appropriately targeted novel or secondary purposed drugs make their way to market in a timely manner.
One key way to increase cost and time efficiencies lies in pharmaceutical companies collaborating and partnering with artificial intelligence (AI) companies that can advise on and ensure the appropriate use of AI and machine learning (ML) in this sector, or even developing these technologies themselves.
The benefits that AI and ML can bring to disease diagnostics and symptom tracking are well understood, with targeted end user wearable apps commonplace. However, given the prevalence of healthcare apps, we will not be focusing on this area of AI in respect of disease and drug development. Instead, our review will consist of two parts: in this first article we will focus on AI and ML in drug development; and in Part Two (to follow in the next edition of Inside Life Sciences) we will cover AI and ML in respect of clinical testing.
AI - Predictive analytic tools and drug discovery
The use of computer simulations in drug development is not a new concept for drug development. However, the addition of predictive analytic tools, including predictive modelling, machine learning and data mining, is the element that has substantially increased the power and effectiveness of computer simulations for drug development.
The overall performance and efficacy of predictive analytic tools rely on two key components – the algorithm, and the data set – discussed in further detail below.
To process data sets, algorithms are required – the more sophisticated they are, the better and greater the analytical value they will provide. In order to ensure algorithms are appropriately sophisticated, they must evolve in line with their analysis of data sets, with ML at the heart of algorithm evolution and development.
Algorithmic ML lends itself to drug discovery due to its ability to extract key information from raw datasets – be it by locating patterns in medical research data, virtual screening, using existing clinical trial data to make efficacy predictions, or de novo drug design.
The most complicated and sophisticated computer systems and algorithms can fall short, unless the data sets that they rely on are appropriately curated, accessible and aggregated. The aversion private entities have traditionally had to sharing clinical datasets publically is increasingly becoming an outdated approach, with private and public entities appreciating the importance of harmonising and centralising data sets through global data sharing platforms.
There are three distinct categories of data sets that can be mined in connection with AI and ML algorithms to enhance and further drug development:
- Molecular Biology Databases – identifying disease targets by way of analysing molecular interactions and microscopy images.
- Structure-Function Databases – creating novel drug leads by way of analysing drug target interactions, molecular structures and structure-function relations.
- Clinical trial Databases – predicting drug responses including drug efficacy and toxicity, and analysing raw datasets to repurpose drugs.
With AI and ML providing a wide array of opportunities with regards to drug development and the repurposing of drugs – whether marketed or otherwise – it is more important than ever that appropriate consents to future research are obtained from study subjects (where appropriate and possible) and that data protection and data privacy remain at the forefront of researchers and pharmaceutical companies' minds when considering collaboration opportunities with AI companies.
A further consideration for companies utilising AI algorithms in connection with drug identification centres on the application of patent law, both in relation to protection of the AI technology used but also in the invention derived using the AI technology. Drug development is an expensive process and patent protection of the newly identified drug provides a recognised incentive for companies to invest in drug development. However, for a patent to be granted, the drug in question must be both inventive and new – the use of algorithms in drug identification presents uncertainty where a drug compound is established using an AI algorithm: does that drug classify as being inventive and novel? If so, who is the inventor: the algorithm or the programmer? One thing is clear, patent law will need to evolve – and quickly – to keep up with the use of AI in drug development.
So what does the future hold?
The opportunities AI and ML present to drug developers are immense. However, as yet, this potential has not been fully realised.
AI is currently supporting disease target identification, drug screening, de novo drug design and clinical predication (to name but a few). However, we are still some way from relying solely on AI and ML findings and analysis associated with drug development. That said, before we know it, AI will not only be performing the majority of stages in drug development, ranging from disease discovery to clinical testing, but also establishing and finding disruptive novel treatments to complex diseases. The question is when.