The application of Artificial Intelligence to the pharmaceutical industry is growing rapidly. Anne Dhulesia and Stephen Roper, partners at L.E.K. Consulting, building on in house insight, consider the use of AI within the pharma industry and its implications for driving forward the discovery of new drugs more quickly, more efficiently and cheaper than traditional approaches. They talk to Nick Herbert
Artificial intelligence (AI) lends itself well to a pharmaceutical industry that has experienced a huge increase in data digitisation over recent years. Its ability to handle vast amounts of data, continually analyse and learn holds great promise for improving the efficiency of drug discovery, repurposing existing drugs and pushing into new areas of science.
It has already made its mark. Take, for instance, the use of AI to speed up development of a Covid-19 vaccine in record time.
There is no universal definition for AI, according to L.E.K., it broadly refers to systems that are able to function with a degree of autonomy and iteratively optimize their processes. Within life sciences, L.E.K. applies the term AI to four major approaches:
- Machine learning processes that analyse input data and then repeatedly optimise their methods based on generated outputs
- Deep learning A machine-learning-based approach that utilizes a logic structure akin to biological neural networks
- Natural language processing A refined automatic speech and written text recognition system that can derive insights from text/speech, going beyond simple reactions to well-stylized user requests
- Robotics and the internet of things Integration of devices to collect, combine and share several types of information
Using these four approaches, AI is set to accelerate, reduce the cost of or replace specific steps in, the drug development process.
‘AI has the potential to analyse large amounts of data and allow decision making in R&D through AI driven algorithms,’ said Dhulesia. ‘It’s optimising R&D processes by removing human error and it’s supplementing and, to some degree replacing, real-life experiments with in silico experiments. AI also enables the discovery of areas of biology to which we previously had no access or no knowledge about by making novel inferences.’
The application of AI in specific steps of the drug discovery value chain, or across various steps, can optimise the process, making drug innovation potentially faster and less expensive.
Currently, around 90% of all clinical drug candidates fail to reach approval, driving the associated costs of drug development to an estimated US$3-5bn. It can also cut the costs of the industry’s R&D spend, which for the largest ten pharmaceutical companies is higher than US$70bn, according to L.E.K.
‘In simple terms it enables us to get drugs to patients more quickly and in a way that is less expensive,’ she said.
There are still challenges to overcome, however, before the full potential of AI within the life sciences and pharmaceutical industry is to be fully realised.
There is widespread recognition among biopharma companies, according to L.E.K., that the R&D process is inefficient and that this inefficiency will continue to grow if left unaddressed. This dynamic is partly driven by the increasingly complex nature of the biology underpinning breakthroughs in the discovery of new molecules and increasing regulatory requirements. But: ‘in contrast to other industries, pharma companies have been relatively slow to adopt AI,’ said Dhulesia.
A reluctance to adopt AI within the pharma sector reflects the conservative nature of the industry and a reluctance to change an R&D process that may have taken years to establish. There are other critical challenges and barriers that need to be overcome, including the belief that pharma has already taken the low hanging fruit in terms of the most attainable targets or drugs and that defining clinical trials is getting harder as patient populations need to be defined more precisely to show the desired effect. The broad reach of AI and its ability to find unexpected correlations and make novel inferences can change that.
‘This is the promise of AI,’ said Roper. ‘Not only to optimise current research methods, but also to go beyond our current understanding of science by supporting the derivation of novel insights from the available data.’
The computational power of AI in analysing massive amounts of data can make previously unknown interconnections between elements or entities much quicker and cheaper than currently achievable, enabling the discovery of new targets and the ability to address new, or acute, diseases.
There are other challenges to overcome.
Data, data everywhere
Sourcing data and cleaning it into a format that is relatively consistent or at least can be analysed is a huge job.
There are two aspects to sourcing data: publicly available, and proprietary. There remain issues around collecting and reformatting data but leveraging this data through AI is already delivering value. And then there are the proprietary data sets that could be sitting within healthcare systems or within specific pharma companies. Sourcing data is more challenging in those instances and requires licencing agreements. Licensing agreements, however, can be complicated and time consuming.
‘Finding high quality data, or data that the AI platform can work with is not always easy and it requires particular talent to develop the algorithms to clean the datasets and interpret the insights,’ said Dhulesia.
That level of knowledge is not always available in house and may be difficult to hire into an organisation. It explains the growing number of partnerships between pharma and AI companies.
‘We see a lot of partnerships between pharma and AI companies,’ said Roper. ‘Because it takes quite a bit of effort to build these capabilities in house for Big Pharma. They all have AI partnerships in place and in most cases multiple partnerships across the drug development chain.’
Nevertheless, outsourcing the AI solution also comes with challenges.
‘It’s hard to identify the right partner or AI solution among all the choice,’ said Dhulesia. ‘And that’s partly due to the current lack of real-world evidence and tangible proofs in terms of clinical outcomes.’
This is becoming increasingly evident as AI derived assets work their way through the stages of clinical development.
‘You also need to be conscious of data privacy requirements,’ she said.
The relative ease of working with non-patient data sets has resulted in AI being more widely adopted in drug discovery than in clinical development, according to L.E.K. The use of patient data is extremely sensitive, and as AI capabilities develop, companies must take the appropriate legal and compliance measures to protect the increasing volume of such data. GDPR compliance in Europe will be particularly important, and failure to comply could have significant reputational and financial consequences.
The way ahead
After several years of experimentation and field-testing, it is time for biopharma players to plan the implementation of AI solutions more broadly in drug development. With the right focus from both AI companies and biopharma to address the barriers, L.E.K. expects that a sizeable proportion of R&D projects will have an AI component within the next five years.