Skip to NavigationSkip to content

Solving the drug development problem

Published on 11/02/19 at 01:06pm

Excitement is building for the potential of AI in just about every field, but how can it address pharma’s unique challenges? Matt Fellows speaks to the not-for-profit organisation the Pistoia Alliance to see where the industry identifies these challenges, and how the technology can help tackle them.

Drug discovery is in crisis. So says Charlotte Deane, Professor of Structural Bioinformatics and Head of the Department of Statistics at Oxford University, in her short article AI and drug discovery: a potential game changer?, echoing the prevailing sentiment in the industry today. The field has always been a challenging one, but with the additional pressures to deliver return on investment and the rocketing costs of the process, the need for effective solutions to quicker, cheaper and more robust drug discovery has perhaps never been felt more strongly.

Today, current estimates place the cost of developing a new drug from the earliest stages and securing its marketing approval at more than $2.5 billion, and considering that around 90% of drugs fail before this point on average, it’s not hard to imagine how expenditures in this area can quickly spiral to astronomical levels. And this rate of failure is even higher is more challenging development areas where data are scarcer – between 2002 and 2012 for instance, the failure rate for Alzheimer’s drugs was 99.6%.

So it’s no wonder why the industry is so desperate to find a reliable, effective solution to galvanising drug discovery efforts, and bringing down costs is a primary driver. On the topic of that ominous $2.5 billion figure, Dr Nick Lynch, a consultant for The Pistoia Alliance, told Pharmafocus: “That number is ever more frightening as it gets larger.” Separating the entire process into drug discovery, drug development and post-launch, Dr Lynch explains: “If you’re looking for where AI could have the biggest impact in terms of cost, it’s likely to be in the drug development part; the clinical trial area is where those costs really ramp up. The use of AI with better modelling could maybe avoid some of the very late failures which really impact that cost model. It could help with running the clinical trials more efficiently, in terms of being able to find the right patients for a particular trial and hopefully compress the timeline that some of the larger trials take, because I think anything that reduces the risk and makes the decision-making happen earlier but also makes it of higher quality will hopefully reduce those numbers in the later stages.”

This alone is benefit enough, but if costs are reduced on the development side, could one reasonably assume that these savings would translate to the price paid by the patient, or whomever the payer may be?

“That’s what the economic model would say,” Dr Lynch responds. “I suppose even now we’re seeing a change in how the industry is rewarded – there is more a pressure to support an outcome-based remuneration. So I think there are factors in play that will make the costs more manageable regardless of AI. In the UK we see the impact that NICE has to manage in trying to look at the cost of new medicines, and I think that outcome-based pricing is happening independent of AI, so in a way, the AI part is hopefully making it more cost-effective.”

How, and why now?

AI has been a hot topic abuzz with excitement in recent years, and not just in pharma and life sciences. And while the benefits the technology promises would seem to justify this hype, how exactly does it function in the frame of drug discovery? Dr Lynch was happy to clarify:

“AI can be thought of as the modelling phase. A model is an ability to represent the real world in a way that you can then start to predict it with new data,” he explained. “You are most likely to have started with some real experimental data and you’ll have built the model, you’ll train it, and then you’ll want to validate it on a different datasets that you haven’t already trained it on, and that will give you an indication of how good your model is. The modellers will work on that to the point where they think the model is good enough for what they’re trying to do. Then they will likely hand it over or it will be used by possibly the same group or even another group, and they might trying to model a particular feature of a molecule – it’s ability to cross the blood-brain barrier, as an example for a neuroscience drug. They will then use that model to help them select which new candidates they should make next, they’ll make them, and that data is then is fed back to the beginning, starting the loop again. It’s very much driven by building a first example and then constantly reiterating it as you discover more experimental data to help shape the effectiveness and the quality of the model that you’re then using.”

While it does seem that the excitement surrounding the decidedly sci-fi technology is reaching fever pitch, you’d be forgiven for thinking it was a new phenomenon in this space. But this isn’t the case, and the industry has been leveraging it in some form or another for decades, as Dr Lynch notes: “I’ve been in the industry long enough to remember when it was perhaps called ‘statistics’. There’s been many luminaries who’ve made very effective use of machine learning for the last 40 years; it’s been a traditional tool for the computational chemist or biologist.”

So why is technology getting everyone excited right now?

“A couple of factors have come to play,” Dr Lynch continues. “The improvement in the deep learning methodology, which has been driven by some of the technology companies and the research – a lot of that was driven initially by being trained on imaging; the sort of imaging recognition that we see in our private lives has driven a lot of this approach in terms of the research, and then that’s had a knock-on effect on the whole area of deep learning and neural networks. Another factor is the improvement in computing power; not only have we got better algorithms, but we can run them quicker, and those two elements in combination means that you can do much quicker cycles – something that could have taken two weeks might take an hour; it makes it so much more usable within a research environment.”

Gauging opinion

The unquestionable promise the technology presents obviously has it poised as the next big thing in drug discovery, but because it hasn’t quite been broken in yet, there are a lot of questions still to be answered as the industry warms to it and begins to understand it. In this spirit, the Pistoia Alliance has been making moves to paint a picture of where attitudes towards the technology currently stand, and where concerns are falling.

“We’ve been running a series of webinars as part of a centre of excellence that we set up, because we felt that in order to tackle AI appropriately, we need to be more coordinated as an industry, and that many of the challenges that exist are probably common to all participants and parties,” Dr Lynch said. During these webinars, industry professionals were asked some key questions, and the findings were published in two reports, one released in January last year, and one in June.

In the first, which included the responses of 374 life science professionals on AI, machine learning (ML) and natural-language processing (NLP), revealed that 44% of respondents were already using or experimenting with AI, with the majority (46%) of AI projects focused in early discovery or preclinical research phases.

As for other applications, 15% were using it in development and clinical (15%), and 8% in imaging analysis (8%). 23% were applying ML for target prediction and repurposing of existing drugs, while 13% were using it for biomarker discovery and 8% for we reusing it patient stratification.

However, it was also discovered that a considerable 11% were not using AI at all, and this figure rose to 27% with NLP and 30% with ML, while 8% said they knew ‘next to nothing’ about AI and deep learning, which the Pistoia Alliance noted highlighted the necessity of cultivating greater understanding of the technology through better education and the sharing of knowledge.

In the second report, which amassed the responses of 229 life science professionals, it was discovered that 69% of companies were using AI, machine learning, deep learning, and chatbots, denoting a growth in these numbers since the same question was asked in September 2017. This suggests we are moving in the right direction, but 72% still said they believed their sector was lagging behind other industries in its development of AI.

This provides a valuable snapshot of the industry as it stands now, but this is just the first steps of a much longer process: the Pistoia Alliance plans to continue conducting regular litmus tests of industry professionals to monitor these trends as they develop in coming years.

We want to follow these trends and hopefully pick up some of the new trends as they develop, because I think it’ll be interesting to see, over the next 18 months, what is the next area that comes forward,” Dr Lynch noted.

The results of the surveys are illuminating, but it is still decidedly early days as far as adoption of the technology goes. I think many companies have been running or at least looking into running what you might call ‘pilots’, but there still exists a hill to climb in moving from a quick pilot to something which is reliable and part of the usual drug discovery process,” Dr Lynch continues. “2017 I think was the awakening of AI pilots, and 2018 was where people were trying to build momentum, and corporate strategies were taking this on board; you had CEOs of companies putting their weight behind an AI strategy – the CEO of Novartis for instance was quite vocal about it.”

Identifying adversity

Of course, while adoption is sure to continue to grow, the path forward will by no means be smooth. Another key reason behind the Pistoia Alliance’s surveys was the drive to identify where the industry’s concerns and anxieties currently stand as they move forward to navigate implementation of the technology.

“By running the surveys, we are trying to identify what areas the community of our members and beyond want us to focus on,” Dr Lynch explained. “We’re actively using it as a way of trying to pick out where the challenges are. People talk a lot about the success of AI, and there are many start-ups which want to build awareness, but for it to be successful we need to tackle those challenges.”

To this end, the findings revealed the difficulties in procuring technical expertise as the most cited challenge for AI (30%) and for ML/NLP (28%). On this, Dr Lynch noted: “Life sciences companies aren’t the only kind of people who require these kinds of skills, as we see in retail and other technology sectors, so there’s perhaps a general dearth of data scientists not just in the US or Europe but across the globe. So life sciences companies are having to compete with perhaps much bigger firms which, because of pay or other factors, mean that it’s hard to always get the right talent that you need. I think that the technology is moving so fast that people need to keep up to date, and companies are looking to hopefully retain the people they have, but also retrain people as well so that they’re not having to constantly recruit new people.”

Beyond this, 26% of those responding to the surveys referenced access to data as a key challenge, while 26% singled out data quality. In regards to ML and NLP projects, these same issues were raised by 26% and 19% respectively.

On this note, Professor Deane paints a particularly elucidating image: “Using AI in drug discovery is often like training an algorithm to recognise pictures of cats when you have no pictures of cats but a relatively small number of out-of-focus, badly annotated pictures of dogs and elephants.”

Dr Lynch had some key insight to share on this: “If you’re wanting to use AI effectively, you need access to both the volume of data and data of good quality; the old axiom of ‘garbage in, garbage out’ is very true when it comes to AI,” he explained. “The science world is made up of data that could have been generated a long time ago, as well as data that we’re generating in 2019, and you’ve got to make the call as to which data you use in the training phase, because you’ve got to perhaps have some degree of scepticism or be cautious about what data you use. There is the concern of ‘how believable is data that was collected five or ten years ago?’, as well as hoping that you can fully understand it, because AI is somewhat of a black box; your ability to fully interrogate the model is somewhat lost because of the way it’s built – it’s multi-layered and quite complicated. So you’ve got to have a lot of trust in your data and your ability to model it, and that’s why the data part of this is so important: because if you train incorrectly with data that you’re not sure about then the risk is that your model is taking you in the wrong direction.

“There’s some good practice that can be applied to data,” he continues. “One of those is what’s called the FAIR principle: making data findable, accessible, interoperable and reusable. There’s quite a growing momentum behind that now because those four themes are important for when an AI algorithm is looking at the data because, unlike a human, they will be analysing data autonomously, and so they don’t always have that human intuition about data right from the beginning. There are ways that we can make progress with that, but it’s going to be slow progress at times just because of the sheer volumes of data that exist, particularly historic health data which could even be held in paper form – our industry doesn’t always have the latest and greatest information techniques for data that existed ten years ago. Our medical records might have been on paper, so how do we learn from those in an effective way?”

Collaboration is key

While the potential of AI as a dynamo for drug discovery is beyond reproach, it’s clear that we are very much in its nascent stages with regards to fully realising it. As Professor Deane notes: “Many challenges remain: none of these methods are accurate to a level that can be used without significant amounts of wet lab experimentation. All of them require human interpretation, and there are still real questions about the generality any of them can or will achieve.

“But AI algorithms and techniques are already changing the way drug discovery is done, and as the algorithms improve, as we gain a better understanding of how to handle and represent the data, and also what data to collect, their benefits can only continue to grow.”

Key to progress in this area will be the expansion of our understanding and knowledge base, and also the sharing of that base among professionals in the industry to solve the problems that are faced by all. With its ongoing approach to interrogating the industry and provoking thought in this area, the Pistoia Alliance look to be on the front lines with regard to cultivating just that.

The organisation said in a statement accompanying the release of its survey findings: “Given how crucial data is to building AI algorithms that reveal meaningful insights, collaboration over data standards, benchmark sets, and data access, will be essential,” while its President, Dr Steve Arlington, commented: “AI has the potential to revolutionise life sciences and healthcare – all the way from early preclinical drug discovery to selecting precision treatments for individual patients. Our survey data shows that while life science professionals are already exploring how AI, ML and NLP can be used – there are clear gaps in the knowledge, data, and skills, which will enable more pharma and biotech companies to achieve tangible results from AI. Impediments to success, such as a lack of industry-wide standards for data format, will need to be addressed, if the potential of AI and ML is to be realised. We urge those in the pharmaceutical, biotechnology and technology industries to explore ways in which they can collaborate now, to find answers to common problems of the future.”

Matt Fellows


Charlotte Deane has shared nice ammount of data about discovering new drugs. He says in book in the present era pharmasotical industries required billion of dollars to discover new drug. I always prefer reading writing services reviews before help in research paper online because it helps to hire the best writer online.

Mission Statement is a leading portal for the pharmaceutical industry, providing industry professionals with pharma news, pharma events, pharma service company listings and pharma jobs,
Site content is produced by our editorial team exclusively for and our industry newspaper Pharmafocus. Service company profiles and listings are taken from our pharmaceutical industry directory, Pharmafile, and presented in a unique Find and Compare format to ensure the most relevant matches