How will AI Disrupt the Pharma Industry?
There is a lot of buzz these days about how artificial intelligence (AI) is going to disrupt the pharmaceutical industry. At one time a mere idea that was the subject of discussions around its possible and practical uses, these technologies are now a reality that is turning pharma into a more productive industry with game-changing applications and many more yet to come.
The face of AI is one recognized by many but trusted only by a few. Pharma companies who have already embraced the technology are reaping the benefits of faster results and more accurate insights, while those who are still watching from the sidelines may soon be at risk. Most of the companies that have started to leverage AI in order to streamline their work have done so in respect to their specific needs. The reason that some established pharmaceutical companies, and even small biotech startups, have embraced the technology with open arms is that it makes previously impossible discoveries possible and verifies hypotheses that could take years if done through traditional methods.
According to a study by Deloitte, 83% of early adopters of cognitive technologies in the general business world state that they have seen moderate to substantial benefits with its use. Machine learning adoption was already high at 58% in 2017 and grew by 5% in 2018.1 On the other hand, many companies still want to see ample evidence of the success of this modern technology before they get ready to embark on an AI-driven journey.
However, AI in pharma and life sciences is not a new concept. The term artificial intelligence was coined by John McCarthy in 1956, although philosophies about the concept started to emerge even earlier. Around the same time, between 1955 and 1956, the first-ever AI program to mimic human skills for problem solving, Logic Theorist, was written. Although even now a substantial number of pharma companies are hesitant about adopting AI, it’s an interesting fact that the industry saw the first instance of it in the mid 1960s with project DENDRAL. The program this project produced studied hypothesis formation and discovery in life sciences. It helped organic chemists identify unknown molecules by analyzing their mass spectra and using knowledge of chemistry.2 This was the first time that AI automated problem-solving and decision-making in the field of life sciences.
But what is artificial intelligence?
As much excitement as there is about this innovative technology, it comes with a hint of trepidation on the part of some around every possible and practical output it can provide. The word cognitive in the definition of the technology tends to make people skeptical about its use. However, in simple terms, artificial intelligence is nothing more than a program that uses knowledge it was previously trained on to find patterns that are not explicit to humans. It is fundamentally important to understand that artificial intelligence cannot do anything a human cannot do given unlimited knowledge and an unlimited amount of time. But time is of the essence, and patients’ lives depend on it.
Biomedical, patient, and medical data are increasing at a pace impossible for scientists and researchers to keep up with in order to make use of it. For instance, in immunology, a new paper is published every half hour. If a researcher takes 1 hour to annotate a paper and 2.5 hours to perform manual research, it will be impossible to stay up to date with the latest information. Densen, 2015, estimated that medical data will double every 73 days by the year 2020.3 Without AI, it will be impossible to collate and curate such a huge volume of data for generating useful insights from it. AI can accelerate decision-making processes that may take weeks or months otherwise.
Another word that creates skepticism around AI is automation. Automating a process can by definition be as simple as using any machine. If we apply this definition to AI, it means giving the program an input and a direction, and receiving an output. Obviously, AI can solve complex problems, thanks to data scientists and advanced algorithms. An AI program collects the data, understands it, reads the patterns (even those previously undiscovered by humans afters years of research), and gives insights in much less time than it takes via manual intervention. Automation is what makes AI essential–it can perform these tasks much faster than humans and deliver results in real time.
So what exactly can AI do for pharma and life sciences?
In 2007, a robot made the first scientific discovery using AI. The robot, named Adam, identified the function of 19 yeast genes. When researchers at Cambridge and Aberystwyth, UK, tested Adam’s hypotheses, only one was wrong. Not only were 18 hypotheses accurate, but half of them were previously unknown.4 In 2018, it was reported that an advanced robot named Eve helped researchers in antimalarial drug development discover the enzyme that triclosan targets, DHFR. This enzyme is also targeted by pyrimethamine, a drug for which the malaria parasite has developed some resistance. The hope is that triclosan, a commonly found toothpaste ingredient, will be the basis of more effective treatment against malaria.5
Before the use of artificial intelligence in the industry, scientists spent most of their time on lower cognitive tasks, such as collecting the information that may be useful for decision-making, for reaching a hypothesis, or for coming up with an important and accurate theory. Researchers spent weeks doing what AI can do in hours before a useful insight could be generated. Nowadays, with the adoption of AI and big data analytics tools by the pharmaceutical industry, generating hypotheses, validating theories, finding unexplored targets and biomarkers, as well as discovering new connections is faster and easier.
AI technologies enable pharma in multiple ways. With computer vision, large volumes of publications and theses can be analyzed in minutes; with network analysis, complex connections between drugs, pathways, genes, and diseases can be visualized; with image-recognition techniques, medical images can be read to generate accurate diagnoses and hypotheses about different indications; with sentiment analysis, unforeseen side effects of interventions can be mapped out; with machine learning, personalized medicine can finally be possible.
In 2018, an AI program using deep-learning algorithms (trained on more than 100,000 images of dangerous and benign skin lesions) accurately identified 95% of skin cancers from a set of images compared with 86.6% accuracy from 58 dermatologists who examined the same images.6 The program will be useful in detecting early-stage skin cancers, thereby saving millions of lives. If we can create such useful tools with the help of AI, its adoption should be strongly encouraged. By leveraging AI, not only are early diagnosis and new discoveries possible but also are a significant reduction in the timeline for drug development and increased market opportunities achievable. In the end, these new technologies will help accelerate patient access to safe and effective therapies.
With the help of AI, pharma and life sciences industries can enable faster, more accurate real-time insights for:
- Understanding the market scenario, trending topics, and unmet needs of patients
- Visualizing vast volumes of medical, patient, and scientific data
- Better streamlining of processes and collaboration among researchers, key leaders
- In silico validation of compounds, identification of biomarkers, etc.
- Enabling access to deep data and minimizing manual efforts
- Generating hypotheses and validating novel approaches
- Discovering known and unknown connections for drug repurposing
- Staying up to date with guidelines, regulatory, and patent landscape
- Getting real-time access to life sciences and medical data
Despite favorable applications of artificial intelligence in the pharmaceutical and life sciences industries, not all companies have completely embraced the cutting-edge technology. Accenture estimated potential annual savings of $150 billion for the US healthcare economy alone by 2026, using AI-based clinical health applications. Additionally, the study observed that approximately 20% of untapped clinical demand will be addressable using AI.7
The pharma industry faces high R&D spends, long drug development timelines, and 9 out of 10 failed clinical trials.8 In the face of these challenges, even small enhancements are worth pursuing. Establishing a strong AI strategy may be the breakthrough disruptor the industry is waiting for in the 21st century.
References: 1. State of AI in the enterprise, 2nd edition. Deloitte website. https://www2.deloitte.com/insights/us/en/focus/cognitive-technologies/state-of-ai-and-intelligent-automation-in-business-survey.html. Accessed May 15, 2019.
2. The history of artificial intelligence: 25th anniversary of the DENDRAL project. Stanford Universities Libraries website. https://exhibits.stanford.edu/ai/catalog?f%5Btopic_facet%5D%5B%5D=DENDRAL. Accessed May 15 2019.
3. Densen P. Challenges and opportunities facing medical education. Peter Densen. Trans Am Clin Climatol Assoc. .2011;122:48–58.
4. Sparkes A, Aubrey W, Byrne E, et al. Towards robot scientists for autonomous scientific discovery. Autom Exp. https://aejournal.biomedcentral.com/articles/10.1186/1759-4499-2-1. Accessed May 15, 2019.
5. Scammell R. AI robot aids scientists in malaria discovery. Drug Development Technology website.
https://www.drugdevelopment-technology.com/news/ai-robot-aids-scientists-malaria-discovery. Accessed May 15, 2019.
6. England R. AI outperforms human doctors in spotting skin cancer. Engadget website.
https://www.engadget.com/2018/05/29/ai-outperforms-human-doctors-in-spotting-skin-cancer/. Accessed May 15, 2019.
7. Artificial intelligence: Healthcare’s new nervous system. Accenture website.
https://www.accenture.com/t20171215T032059Z__w__/us-en/_acnmedia/PDF-49/Accenture-Health-Artificial-Intelligence.pdf. Accessed May 15, 2019.
8. Sullivan T. A tough road: cost to develop one new drug is $2.6 billion; approval rate for drugs entering market is less than 12%. Policy & Medicine website. https://www.policymed.com/2014/12/a-tough-road-cost-to-develop-one-new-drug-is-26-billion-approval-rate-for-drugs-entering-clinical-de.html. Updated March 21, 2019. Accessed April 2, 2019.