Needle in a haystack! What tech advances are shaping drug development?

Drug research and development is often likened to finding a needle in a haystack with multiple targets and many potential variables with which they can interact. So, what technological advancements are impacting development timelines and how can they be applied on a wider scale? Paul Denny-Gouldson, vice president strategic solutions, IDBS, goes into more detail…

Many people ask: “What new and emerging technologies will shape the future of drug research, development and delivery?”

From research and development (R&D) through to manufacturing the final drug product, the range of processes that take place before medicines are administered to a patient are extensive, expensive and time consuming. Yet, early data suggests these timelines are beginning to shorten and, in some cases, costs are beginning to lessen. So, what is causing this trend — and how can it be replicated?

Drug research

The problem in drug research is often associated with the ‘finding the needle in a haystack’ conundrum. There are so many ‘targets’ within the human physiology system — such as receptors, enzymes, proteins, pathways — and so many potential variables that can interact with these targets, that both the intended and unintended effects give rise to the needle in a haystack simile. With so many variables it can become almost impossible to identify how a drug works and what pathway it is affecting.

The development of artificial intelligence (AI), in all its possible guises, can go a long way towards solving the problem. There are currently several endeavours that are seeking to exploit the massive computing and pattern analysis capabilities of modern AI platforms. These also include the analysis of historical data of potential candidate compounds that were previously ‘canned’, with the aim of repurposing them for other diseases.

Further exploits are also sighted towards the analysis of complex data sets to elicit new information about a given drug target or pathway — where connecting massive disparate data sources, both internal and external, is a key requisite for success. Other areas of interest lie in the area simulation of drug properties and effects on the human body. Whilst this has been done for many years, the type and size of data feeding the simulations are growing rapidly.

Lastly, we are also seeing emerging trends where scientists are being supported in their daily work with augmented intelligence. This is where the decision is still made by the scientists, but the data and support for that decision is as automated as possible. The request for an AI platform to flag information, as and when it is needed, is a natural extension of AI’s offering, as it sits squarely in the arena where AI can have the greatest impact.

Drug development

The problem faced by many R&D organisations when developing new drugs is in finding the shortest, quickest and most cost-effective path to the finished product. This is due to the extremely stringent regulations and quality directives that drug development processes must adhere to.

Yet the commercial pressure on companies is always there — whether improving competitive advantage, improving market share or being first to a new market — each drives behaviour. A single day’s delay of a drug to market costs a company an average of $1 million in lost revenue.

Automating decisions and aggregating data that is relevant to decisions is an area of great interest to the application of AI technology, and can go a great way towards solving this problem. Whilst AI cannot offer a silver bullet or a single answer, automated tools can greatly improve the speed of data aggregation, greatly improve the surfacing of the relevant data to the right person or group at the right time, and thus speed up the discovery process.

The automated control of manufacturing processes is also an area of great potential for AI, given its ability to monitor many variables and interrogate data at a speed that is not feasible for humans. All this leads to earlier identification of either unwanted, or desired, trends in the process and a self-learning system that can optimise and make decisions based on inputs. This operates in the same way as autonomous cars — via machine learning and pattern identification — but the environment in the pharma process space is not as well defined as the automotive industry, and so may take a little longer to develop.

Drug delivery

Following the R&D phase, the next challenge in the process is the need to get the right drug to the right patient at the right time. This, like other issues earlier in the process, is a multi-layered problem, where there are many variables at play. The advent of ‘-omic’ medicine (such as genomics, proteomics, metabolomics), alongside our continued ability to look at the fundamental and underlying reasons why diseases manifest, is going a long way towards improving the process, allowing scientists to tailor the drugs that patients are given.

However, the amount of data and its complexity means that AI and associated methods are critical to both understanding the disease (related to the first discovery problem statement) and using it to make treatment decisions. Coupled with this data deluge comes the added complexity of individual patients — a drug that might work at one time for a given patient/disease might not work in another instance.

Another area of clinical research that is well underway in identifying delivery issues is the monitoring of adverse events once a drug is released. Here AI comes into its own, as it can help analyse the terminologies used to describe events and learn how to categorise those based on previous understanding and training undertaken by scientists. All this can be done on both social media and hospital data (with the correct security in place) allowing drug companies to get an almost real-time view on the status and performance of a drug as it is administered to patients.

Some work still to do

The implementation of these new technologies described, such as deep learning, machine learning and augmented intelligence — all delivered via the cloud — is allowing greater global collaboration, faster research decisions and more effective crunching of increasing amounts of data. All of this is going a long way to finding hard to identify problems within complex ‘haystacks’.

The trick now, as these technologies continue to develop, is to really understand what they can offer outside of the small, niche areas of application they currently serve. Without such general potential applicability, it will become necessary to reign in the industry’s expectations of impact, and tailor them to be far more realistic, as there is very rarely a silver bullet for problems.