Artificial intelligence (AI) is currently one of the biggest buzz words within technology innovation. In the simplest terms, AI is a branch of computer science that involves training computers to perform tasks such as learning and problem solving, typically thought of as requiring intelligence. For example, AI could involve a computer learning through trial and error to identify complex patterns in events or play strategic games. Tasks within AI are frequently divided into two types: supervised and unsupervised learning. In supervised learning, the desired outcome of the task is known. Identifying a cat in a picture is one example. In unsupervised learning, the desired outcome of the task is unknown, and the goal is generally to learn some previously unknown underlying structure within the data. An example of unsupervised learning could involve identifying clusters of similar genomic sequences linked to tumors.

AI permeates almost everything around us today, influencing digital assistants on smartphones, navigation, targeted advertisements online, music, and even the suggestions offered to users on streaming services. AI and machine learning are notably becoming increasingly adopted for numerous applications in healthcare.

While advances in science and technology have enhanced patient treatment and survival, there is considerable room for improvement. With the growing amount of data and disease complexity, AI can be beneficial, as it can go beyond the capabilities of a human brain to observe and learn on a larger scale. In healthcare, AI has several applications not limited to recruiting eligible individuals for clinical trials, using electronic medical records to forecast patient outcomes, deploying medical imaging for diagnostics, and enhancing drug discovery.

Why is AI So Popular Now?

Although AI has recently experienced a surge in popularity, it is not a new concept, especially in healthcare. In the 1960s, DENDRAL, the first example of what is considered an expert system—a computer program designed to encode and then mimic the decision making of a human expert—was developed at Stanford University to derive compound structures from mass spectrometry and experimental data.  A number of ventures subsequently drew inspiration from DENDRAL. MYCIN, which was applied to handle bacterial infections, recommends the most appropriate antibiotics with weight-adjusted doses. CADUCEUS broadened these applications to include other ailments. AI’s implementation has augmented due to advances in supercomputing, decline in computing and data storage costs, and the ease and improvements in data gathering. With improved hardware and access to larger datasets, the software and science of AI continually progress, leading to increased optimism, adoption, and shunted global financial investment.

Setting the Health-AI Stage, So to Speak

A simple web search on AI in healthcare yields several pharmaceutical companies and AI start-ups with diverse applications. However, a simple conversation with friends or colleagues lacking specialized knowledge of AI immediately exposes many gaps and misconceptions in people’s understanding of AI. This confusion is an issue that Annastasiah Mhaka, PhD, Senior Advisor at Adjuvant Partners, is working to combat by co-founding the Alliance of Artificial Intelligence in Healthcare (AAIH). Dr. Mhaka, AAIH’s President, is excited to be at the forefront of an international multi-stakeholder organization that impacts where “life sciences marries technology.” She recently discussed AAIH’s establishment on Talking Precision Medicine, a podcast hosted by Rafael Rosengarten, CEO and co-founder of Genialis, also an AAIH founding member.

The AAIH is a global organization with 22 founding organizations all pursuing different aspects of AI in healthcare. Inspired by the success of the Alliance for Regenerative Medicine—an international organization with members from different sectors involved in and dedicated to research, advocacy, and commercialization of gene therapies, cell therapies, and tissue engineering—the AAIH was developed with the goals of driving awareness of AI in healthcare, advocating for regulatory guidelines and standards for its implementation, fostering cross-industry collaborations, and promoting capital formation.  AAIH member organizations develop or utilize AI in biomedical R&D and clinical applications, and include growth phase start-ups, biopharma, diagnostics and device manufacturers, and research institutions. Leaders in each of these founding organizations serve on the board of the AAIH to work towards its initiatives, cultivate solutions to combat challenges in the field of AI in healthcare, and in turn implement AI in their respective fields for sustainable and accessible data-driven discoveries that will improve patient quality of life.

Oh, the Places AI Could Go, Drug Development-wise

With the advances in AI and the influx of biomedical data, there is potential to significantly strengthen the future of clinical medicine and the pharmaceutical industry. Traditional drug discovery is a decade long process with high failure rates. Dr. Mhaka is optimistic that AI can “help increase the probability of success, [and] decrease the number of experiments and costs associated with drug development while delivering more effective drugs.” There are a number of companies using AI for drug discovery (Table 1).

Table 1: 12 Companies Implementing AI and Machine Learning in Drug Discovery

Numerate is a San Francisco based company founded in 2007. It is one of the founding members of the AAIH, and its co-founder and Chief Technology Officer Dr. Brandon Allgood serves as AAIH’s Vice Chair. Numerate is one of the longest standing AI drug discovery companies. It also has a diverse team with over 80 years of experience in drug discovery. To tackle the “multiparameter optimization problem” of drug discovery, an issue appropriately termed by Dr. Allgood, Numerate uses AI to identify small molecule candidates. Numerate’s platform trains high dimensional proprietary learning systems on medicinal chemistry and biological data, which is “medium data not big data,” humorously termed by Dr. Allgood.  This AI-driven platform includes Numerate’s patented ranking models that prioritize compounds based on molecular criteria and implement absorption, distribution, metabolism and excretion (ADME) and toxicity model predictions. Furthermore, this platform also predicts biological effects using another patented method granted to Numerate.  This validated platform discovers, profiles, and prioritizes hits from a virtual screen of commercial compound libraries, thereby decreasing the need of synthesizing and testing unnecessary compounds and reducing time and cost.

Numerate funds its operations through investments, public funding such as grants and contracts, and its strategic alliances. Its success has led the company to secure contracts and establish collaborations with companies such as Takeda Pharmaceutical Company Limited, Lundbeck and Servier through which Numerate leverages its platform to identify candidates for multiple therapeutic areas. Along with Glaxo Smith Kline, Lawrence Livermore National Laboratory, Frederick National Laboratory for Cancer Research, and the University of California, San Francisco, Numerate is also a member of the Accelerating Therapeutics for Opportunities in Medicine (ATOM) consortium. ATOM aims to develop a platform for drug discovery that can shrink the process of drug target to patient delivery to a year by integrating “high performance computing, diverse biological data, and emerging biotechnologies” according to its website.

Nuritas is a Dublin based company founded in 2014 by Dr. Nora Khaldi. Like Numerate, it is a founding member of the AAIH and Dr. Chantelle Kiernan, Director of Lifesciences, serves as the co-chair for the AAIH Federal Engagement and Regulatory Affairs (FERAC) Committee. According to Dr. Kiernan, Nuritas is a first of its kind which “combine[s] science, AI, and nature to discover completely novel bioactive peptides with health benefits in strategies of both prevention and cure.” The company uses AI to interrogate natural and safe sources of matter, such as food, to identify completely novel drug backbones that have a lower probability of clinical failure due to safety and toxicity compared with certain other drug formats. Nuritas’s AI platform includes three phases: “Target,” “Predict,” and “Unlock”. In the “Target” phase, researchers retrieve all public domain data relevant to a specified area of interest or disease. This public domain information is supplemented with completely novel scientific data on peptides generated in house. Unique to Nuritas is a proprietary machine-readable natural peptide database containing information on billions of peptides, explicitly in silico predicted peptides and over 1 million mass spectrometry analyzed and novel peptides, collated over the last 5 years. In the “Predict” phase, the AI platform interrogates both the public domain and their proprietary database “to discover novel peptides capable of modulating the target or disease area of interest”. Finally, in the “Unlock” phase, the AI platform identifies the “unique enzymology” necessary for isolating desired peptides from identified source materials.

Nuritas is the first company in the world to have a commercially validated AI platform, following the successful launch of its AI discovered peptide that reduces circulating TNF-a, in collaboration with BASF in 2018 (PeptAIde^TM). As part of its Preventative Strategy, natural unmodified peptides are currently being evaluated in the clinic to assess their impact in pre-diabetes. With respect to its Therapeutics Strategy, Nuritas has identified multiple novel peptides from natural sources capable of modulating disease. Such peptides serve as the basis for subsequent synthetic strategies that modify and optimize their efficacy and drug like properties.

According to Dr. Kiernan, Nuritas generates revenue by licensing its pipeline candidates to customers for “further development and commercialization.” Additionally, through its “On-Demand” service, Nuritas works closely with partners to discover novel bioactive peptides in areas specified by these partners. Nuritas discovers and validates these peptides, leaving the role of clinical evaluation and commercialization to the partner (Therapeutics Strategy). Dr. Kiernan notes how “in both cases, our partner takes the discoveries to market.” Leveraging its AI platform, Nuritas has maintained successful partnerships thus far with Nestle, Pharmavite, and BASF and is presently in multiple late stage negotiations with some of the world’s largest pharmaceutical companies.

The Fine Print

While AI and machine learning can provide new solutions to advancing discoveries in healthcare, challenges still exist. Dr. Uwe Klein, Head of Biology at Numerate states that wet lab work and the necessity for validation add to timelines and cost. In addition, extensive time is required to synthesize compounds, distribute these compounds to labs, and conduct in vitro and in vivo tests when necessary. Dr. Allgood confirms that “a single [wet lab] data point can cost $1000.” He also notes that data poses a challenge for most applications in the drug discovery space. He states that on most occasions, acquiring good quality data is difficult especially from the public domain and from within pharmaceutical companies, as biological data is always “noisy, and biased.” The reality is that a decent amount of time is spent preparing data—cleaning, normalizing, and annotating—for appropriate representation to the respective AI algorithm.

Another challenge is the talent required for AI in healthcare. Dr. Kiernan highlights that “finding [people with] the right skills will be a growing issue, [and] as more industries move towards digitalization, increased pressures are being placed on the global talent pool.” Dr. Kiernan further suggests that more needs to be done on the national level to support skill development, which “ranges from new graduates to upskilling programs for existing professionals.”

Accompanying the necessity to build skills in AI, a critical issue to be addressed are the gaps in knowledge and lack of standardization. Precisely, more education is essential to aid the public’s understanding of AI’s capabilities and limitations and to reassure them that it is a tool that still requires human involvement. As Dr. Kiernan states, “we are still very much in control of AI.” There is promise that with the establishment of the AAIH, the industry will become more knowledgeable about AI and can observe its practical and ethical execution. In fact, in collaboration with other AAIH members, Dr. Allgood is preparing a White Paper to initiate the organizations’ goal of educating and driving awareness. The AAIH intends to have this document as a guide to define terms, present thoughts, and create a foundation for its future work.

Is there a future for AI in healthcare? Time will tell. Although, IBM Watson, a key player in AI in research and discovery, halted its sales and development of its drug discovery product for poor financial performance, there is still steady promise for AI’s future. Major technology companies like Google, Microsoft and Amazon are significantly investing in healthcare, and new applications are continuously introduced. In fact, the US Food and Drug Administration’s latest push to develop guidelines for AI in healthcare signifies its recognition that this approach is lasting. There are certainly a great deal of emerging companies in the AI in healthcare space with unique angles, and there is room for new applications throughout the drug discovery pipeline. For one, Dr. Klein is excited about the possibilities on the diagnostic end for the identification of subtleties humans cannot detect and patient stratification for treatments using numerous measurable factors.

Moving forward, the field of AI in healthcare will benefit from (1) education to drive awareness of AI and to train individuals to bridge specialized knowledge in the science relevant to applications with the AI technology, (2) standardization of AI to allow for reproducible and ethical implementation, (3) cross-industry collaborations, and (4) enhanced data access and management to promote larger datasets that are better curated, and machine-readable. Ultimately, AI is not a magic bullet, but it could be highly instrumental if deployed properly.

Stella Belonwu, B.S. is a Pharmaceutical Sciences and Pharmacogenomics PhD Candidate at the University of California, San Francisco.