Promises & challenges: AI in pharmaceutical drug development
The field of AI has revolutionised the way we do things – from the development of self-driving cars to breakthroughs in natural language processing (NLP) and computer vision, AI grows ever-more pervasive.
As the AI ‘gold rush’ rages on, in what seems to be an unrelenting fashion, it comes as no surprise that, increasingly, pharma companies are utilising AI for drug discovery and its development capabilities.
We explore some of the advantages, as well the drawbacks.
Mining through the data
Vast data sets are synonymous with the healthcare industry. Biotech companies are faced with the momentous task of mining through troves of patient data, including health records, chemical compounds, genetic information, and medical imaging, in search of the next gem that just might be a new, breakthrough drug.
AI algorithms can be used to quickly and accurately analyse this data, identify patterns, and trends, and also make predictions. This can help to speed up the process of finding new drugs, reducing the costs associated with traditional drug discovery methods.
Paul Agapow, Leader in Biomedical Data Science & Machine Learning at GSK, says, “It’s a data-rich process. It's the right optimisation and, if you could just lean on part of this process, make it 1% more efficient and so forth, there could be great benefits, and great savings there.”
GSK has made some major strides within the field, recently entering into a three-year collaboration agreement that provides GSK with access to Tempus’ AI-enabled platform, including its library of de-identified patient data. Through its leading AI/ML capabilities, GSK will work together with Tempus to improve clinical trial design, speed up enrolment, and identify drug targets. This will contribute to GSK’s R&D success rate and provide patients with more personalised treatments, faster.
The problems with mass data
The process does not come without its pitfalls – specifically, the ‘big data paradigm’.
AI algorithms rely on high-quality data to make accurate predictions and decisions. However, the data used in drug development is often incomplete, inconsistent, or biassed. This can lead to inaccurate or unreliable results from AI algorithms, which can, in turn, impede the drug development process.
One of the major difficulties is the fact that complex human biology is context-dependent and that data can be hard to attain.
“We know that, for almost all of these ML models that have come up in various contexts, if we apply them to real patients, they fall apart,” states Agapow.
“There were a couple of papers during the pandemic that concentrated on the point of the ML model, how we would diagnose, and outcomes. The result was that almost none of these models actually won. They were interesting exercises, they had good metrics, but they were not actually very good at treating real people because of the data situation.”
AI in drug identification and administration
AI has the capabilities to reduce human error, identify new drug candidates, and carry out treatment plans efficiently.
In particular, ML can aid healthcare professionals in dosage optimisation – utilising algorithms to analyse data on a patient's medical history, genetics, and response to treatment to determine the optimal dosage of a drug. This can help improve the treatment's efficacy and reduce the risk of side effects.
Further to this, there are opportunities for treatment plans and personalised medicine. AI algorithms can be used to analyse data on a patient's genetics and medical history to create personalised treatment plans. This can help to improve the effectiveness of treatment and reduce the risk of side effects.
We must move beyond seeing drug administration as just a simple task, and instead, one that is complex with many nuances, depending on the patient’s needs and requirements.
“A drug is not just a drug. There is this whole ecosystem around it - the schedule, how it's administered correctly and so forth. This is a great opportunity for ML and AI,” explains Agapow.
“How can we use the software as a medical device to increase the accuracy of administering drugs? So, using software ML models to diagnose people and monitor how they are doing following a course of treatment so we can intervene at the right time – we can prioritise patients.”
Challenges surrounding AI regulation
The increased application of AI within the healthcare industry – and more specifically, drug development – will depend heavily on laws surrounding regulations. AI-related risk mitigation could spell avoiding high-risk consequences for companies.
FDA officials have acknowledged that the rapid pace of innovation in the digital health field poses a significant challenge for the agency. They say new regulatory frameworks will be essential, to allow the agency to ensure the safety and effectiveness of the devices on the market without unnecessarily slowing progress.
Examples of approaches currently taken by the FDA include routine monitoring of SaMD products by manufacturers to determine when an algorithm change requires FDA review and premarket assessment of SaMD products that require it, among others.
AI use in healthcare has the potential to bring about significant changes in patient outcomes, efficiency, and cutting times in research and development. As a result, stakeholders – such as healthcare providers, software developers, and researchers – are continually exploring and creating new AI-based products, which push the boundaries of the current regulatory framework. Regulatory bodies are working towards addressing these challenges by devising policies that foster innovation while safeguarding public health.
There are, however, many issues that must be addressed to achieve this goal. As these policies are developed, it may also be necessary to consider legislative action to clarify the regulatory ambiguities within the field.