Artificial Intelligence in life sciences: why are we lagging behind in AI adoption?
Compared to other industries, in life sciences putting a new product to market is nothing short of a miracle, as it comes with extremely high costs and risk. It is estimated that getting a drug to market bears a cost in excess of $2 billion, with approximatively 60% linked to clinical development. It takes around 15 years to develop a new potential treatment, with no certainty of it ever getting to patients. The attrition rate is extremely high, with an estimated 15% of potential drugs in development successfully reaching the launch phase.
To reduce the cost and increase the probability of success, life sciences would seem a model industry to benefit from the implementation of artificial intelligence tools — and yet as an industry we are still lagging behind others in taking advantage of the benefits of advanced AI. Why?
Let’s get back to some basics.
Although no single accepted definition of AI exists, we can generally describe it as a collection of technologies that enable machines to perform tasks with human-like levels of intelligence. Those technologies enable computers to 1) perceive/sense the world around us by acquiring text, sounds, images, through computer vision or audio processing, which they will 2)comprehend by analyzing trends and patterns in those elements through machine learning techniques, to finally 3) act accordingly through software or human decision making systems.
These three key technologies bring some amazing solutions which facilitate modern workflows, such as virtual agents for online customer services or speech analytics such as Siri. One of the more advanced current applications of AI is autonomous cars, whereby essential technologies understand the world around the vehicle (e.g. a red light), make sense of it (recognize a stop command) and generate a response (controlled braking), all in real-time.
In his recent book “AI Super Powers” (September 2018), Kai-Fu Lee compares AI to the inception of electricity, with all the changes it brought to the modern world and its evolution. Our 21st century economy has already started seeing the light from AI electricity.
AI inception dates back to 1950, when Alan Turing questioned if a machine can imitate human intelligence, and Marvin Minsk in 1951 successfully computed the behavior of a rat searching for food in a maze. Many “boom and bust” phases of successful AI implementation have occurred since then, most due to the limitations of computer power and the cost of analysis reaching the trillion-dollar level. Innovation in computer technologies have enabled the cost of one gigabyte of data storage to decrease from $0.5M in 1980 to three cents in 2015. Most importantly, the availability of large volumes of data connected with the advent of Internet and cloud computing has fostered rapid expansion of AI technologies. Data is the fuel of 21stcentury AI.
Currently, AI technologies are being used across industries ranging from telecommunications, where virtual assistants are implemented to support bug fixing and predict customer usage patterns, to financial services, where it is estimated that 60% of Wall Street trades are executed by AI tools with little or no human interventions. Energy, automotive, travel, and retail sectors all currently use AI technologies to either improve operations and processes, increase customer satisfaction, forecast maintenance, and otherwise improve return on invested capital.
However, when compared to the widespread adoption in other large industries, little AI technology is employed within the pharmaceutical industry.
A recent survey of life science leaders (Accenture 2018) recognized that 85% of respondents expected AI to facilitate the expansion into new markets, new products and services, new models. 74% foresaw AI to have a complete or significant impact across life sciences.
AI technologies have the possibility to increase ROI within drug development at early phases driving target identification and lead optimization, in clinical development improving patient identification and recruitment and in clinical trial management reducing time, cost, and patient attrition rates. Other industry areas are benefiting from AI tools — disease identification, where by analysis are carried out to get better understanding of specific diseases, drug discovery, in which intelligent algorithms are built to better understand disease processes and treatments, personalized medicine, investigating customized therapies that can be effective for individuals based on data and predictive analytics, and clinical research where recruited patients are better targeted and able to participate in remote monitoring, yielding real-time data access.
In healthcare delivery, algorithms are now capable of detecting the differences in healthy and cancerous tissues; as a result, radiologists are able to supervise algorithms to detect non-healthy tissues.
Despite the investments in AI made within life sciences, there is scarce positive news or precedence showing widespread innovation and return. In fact, April 2019 saw a major setback when IBM ended sales of their Watson machine learning/AI Drug Discovery Services.
Likely impeding the success of AI in life sciences are a few key differences compared between the biopharmaceutical industry with others: The first is the aim to apply AI to endlessly complex human biology.Our bodies are extremely complex — made of over 40 trillion cells working together; it can be argued we have limited understanding of the human organism in health, when our body is working well, so it is extremely difficult to foresee how it is changing when diseased and to understand the causes. We have little control of any of the changes occurring in the body whilst it is going from healthy to sick, so applying technology to support or supplant insufficient knowledge is a high bar indeed.
A second key difference lies in the high level of regulation of life sciences innovations. Most therapeutics cannot be put on the market without consent from country-specific regulatory agencies. In light of AI evolution in healthcare and other industries, the U.S. FDA is looking into speeding up regulatory guidelines on products which incorporate AI technology and algorithms, an evolving topic currently.
A third key element limiting AI expansion in life sciences is data, or the lack of. While life sciences is an industry awash in research and published studies, many disease states currently still lack appropriate, large-scale datasets on affected populations with which to build cognitive algorithms. Data collection is largely limited by the level of analytical research into human organisms’ function, as well as the extremely secretive and competitive environment in which the pharma industry is working. The high cost, risk, and competition linked to the development and commercialization of drugs triggers correspondingly high levels of protectionism in terms of shared knowledge, science, research, and data.
What are the next steps to reap the benefits of AI in life science?
Good quality data are the fuel of AI, and we are now seeing leading pharma organizations partnering with major technology companies or data providers to extend large-scale collaborations. Recent programs between GSK and 23&Me, Sanofi and Google, seek to gain a better understanding of disease at a genetic level, and a better understanding of patient populations, with the aim of improving treatment outcomes. More recently, Novartis announced a collaboration with Microsoft to leverage data and artificial intelligence to transform how medicines are discovered, developed and commercialized, speeding up the process and risk associated to the development of a new treatment for patients globally.
Each side of the partnership will bring the expertise in their discipline, one being science and the other one technology, to implement AI as efficiently as possible for the benefit of global health.
Implementation of AI tools within GlobalReach BI has been a key part of our 2019 strategy, with the aim of facilitating decision making for our pharmaceutical partners when assessing potential risks and partnerships for new assets and bringing a new level of objectivity to potential outcomes. GlobalReach’s current AI model enables assessment of key features driving success in clinical trials, based on a test set currently in breast cancer. Mining the clinical trial results available in this area and analyzing each parameter relevant to each phase, yields a check of the probability of current trials moving along the development process successfully. We are no longer limited by a single company’s experimental data, as we are able to collate industry wide data to generate much more specific and focused predictions than was previously possible.
GRBI’s Predictive Risk & Evaluation Platform is now ready for viewing, you can reach me at Yamina.hakem@GlobalReachBI.com
To conclude, our experience at GlobalReach BI has shown how it is paramount to understand the business questions which can benefit from fast analytics employing large data volumes. Focus is essential in early stage implementation, so we are starting small, with targeted use cases, but we are starting now. Data being the fuel of any AI Project, enough data (and clean data) need to be available to draw enough conclusions and generate clinical recommendations. AI is still in its infancy within life sciences; as electricity brings light, we aim to bring light to our partners in life sciences.