Artificial intelligence in drug discovery: an evolution, not a revolution
The following article is an opinion piece written by Andrew Radin. The views and opinions expressed in this article are those of the author and do not necessarily reflect the official position of Technology Networks.
Artificial intelligence (AI) is being increasingly adopted in the pharmaceutical industry, creating both excitement and questions about its potential and long-term success. In recent years, a plethora of companies – ranging from big pharma to startups – have defined AI as a panacea that will revolutionize the industry. While the idea of AI as such is appealing, it is not realistic.
Investors have largely embraced this hype by pouring unprecedented funding into AI-focused startups. However, the immediate expectation of new therapies for refractory diseases has not yet been met. As such, we are seeing a wave of devaluations, dwindling investment and strong disappointment with the industry.
There is a reason for this. We simply cannot deploy a single machine to find new treatments against complex human biology. The science of drug discovery is rigorous and as such our expectations need to be realigned. The practical implementation of AI and the way we think about it associated with drug discovery is going to be an evolution, not a revolution. The place of AI in discovery is a complex relationship that needs to be carefully managed, it is by no means a panacea.
AI is a branch of computing designed to mimic the way the human brain solves problems and makes decisions. It has been around for almost 100 years and the use of AI is not new in the history of pharmaceutical innovation. The sophistication of drug discovery has evolved over the decades and, in fact, AI has been used to help support this evolution, although it is not widely discussed. A classic example is using AI models to help determine relationships between the structural properties of chemical compounds and biological activity. They are essential to drug discovery and help scientists better predict how a drug candidate will act in the body. Although their predictions are limited by model constraints, they have introduced great efficiency into the drug discovery process, allowing scientists to focus on potential drugs that have an increased chance of fighting a particular disease.
Today, however, we are trying to solve much more complex diseases and to fight them with more precision, safety and efficiency than previous treatments. Fortunately, we are now in an era containing a wealth of data on human biology in addition to the ability to analyze large amounts of this data through inexpensive and powerful technology. The potential for AI to treat these complex diseases has increased dramatically, with the caveat that therefore has difficulty in finding treatments and cures.
We can now construct a set virtual world around drug discovery, including in silico models that simulate human disease using large amounts of genomic, phenotypic and chemical data. These data can be consulted freely and analyzed at little cost. We can use computational methods and algorithms to identify features of disease that discovery methods typically miss due to their reliance on a single predefined hypothesis. We can evaluate potential treatments against multiple targets at the same time. As humans, we can only do one thing at a time. The AI fills that gap for us, but it still needs us to guide it along the way.
We can increase the speed at which drugs are introduced into preclinical testing by using AI to truncate the steps needed to get started live test. We can cross libraries of potential compounds against disease targets at lightning speed. We can now better predict the viability of these compounds relative to safety and efficacy markers. This level of progress would take years using traditional methods, but by integrating technology, for just one or two compounds, we can do it all in weeks.
Ongoing investments in AI are paying off. However, unrealistic expectations unintentionally create barriers to even wider adoption. Several AI companies have identified new treatments against new disease targets that have great potential to treat previously incurable diseases, including lupus, glioblastoma, aggressive cancers and fibrotic diseases. The fact that we can use AI to accelerate the search for these potential new treatments is a huge success – it creates a pipeline of new drugs that could soon change the way we treat disease.
AI is already impacting drug discovery in new and previously unimaginable ways. But it all depends on how we judge success. If it’s a machine that alone cures a complex disease, then we’ll never be successful.
The potential of AI multiplies when paired with better education, because with a better understanding of possibilities and expectations, more adoption will occur. The more companies we have improving drug discovery with AI, the more treatments we will find over time. However, these candidates still have to endure years of clinical research and be proven to be safe and effective in humans. While we may have changed the timeline aggressively with the right application of AI, we still have a roadmap to follow that will take years and require rigorous scientific work.
It will be an evolving science. The strongest players will continue to generate a steady stream of results, even if they arrive more slowly and with less fanfare than founders, investors and the media had hoped. This constant stream of empirical evidence will lead to a new appreciation of AI. The one where true value is delivered.
Any scientist working in an R&D lab will tell you that they have harnessed all available technology to its highest potential in the spirit of treating disease and improving lives. For us to say that AI will revolutionize their work is a disservice to all the innovation that has come before us. We need to continue to treat it as an evolution that will take place over time and understand how far it has come.
The fact remains that AI has been in use for decades, evolving with the availability of more powerful computing power and the availability of data. This will continue and we will discover more breakthroughs as a result. These breakthroughs won’t happen overnight, but they will happen.
About the Author:
Andrew A. Radin is co-founder and CEO of Aria Pharmaceuticals. Andrew created the company’s first drug development algorithms while studying biomedical informatics at Stanford University in 2014. Since co-founding Aria, Andrew has been named an Emerging Pharma Leader by Pharma magazine Executive, was invited to give a TEDMED talk and was named one of the top 100 leaders in AI by Deep Knowledge Analytics. In addition to his CEO duties at Aria, Andrew is an advisor to Stanford University’s SPARK drug development and Stanford University’s StartX startup accelerator programs. Prior to co-founding Aria, Andrew served as CTO at several successful internet startups. His past projects have impacted tens of millions in telephony systems, ad networks, video games, and geo-mapping systems. Andrew studied biomedical informatics in the SCPD graduate program at Stanford University and holds an MSc and BSc in Computer Science from Rochester Institute of Technology.