HONG KONG – The use of artificial intelligence in drug discovery was back in the spotlight on the last day of the 2020 KoNECT-MOHW-MFDS International Conference, with drug developers pointing out both challenges and possible solutions.

Nicholas Kenny, chief scientific officer at Syneos Health Inc., said the biopharmaceutical industry “tends to suffer from the complexity and inefficiency inflicted by years of being very document-centric, and not really understanding the data available in a much smarter way.”

He also said that biopharma R&D groups are making significant investments into technologies such as AI and machine learning for drug discovery and patient identification. A Pistoia Alliance survey published in Scientific Computing World’s 2018 Feb-Mar edition indicated that 44% of life science professionals are using or experimenting with AI, and 94% expect an increase in the use of machine learning within two years.

Kenny’s statement was echoed by Youngshin Kwak, senior research investigator at LG Chem Ltd. Kwak quoted Microsoft CEO Satya Nadella, who told a November 2019 congress attended by both men that AI “represents one of technology’s most important priorities, and health care is perhaps AI’s most urgent application.”

Kwak agreed, saying the use of AI could improve the efficacy and success rate of a process that he described as “much more complicated than rocket science.” He also compared the process of drug discovery with sending a probe to Mars, in that both events face consistent hurdles with a high chance of failure.

The use of AI in drug development is not a new phenomenon; a system called computer-aided drug design (CADD) has been used in drug discovery for more than 20 years. Although it was widely used in all areas of drug development, starting with target identification but excluding clinical trials, the big gap between CADD and labs actually developing new drugs rendered its contribution to new drug discovery ultimately minimal.

The drug discovery process also involves the generation of an enormous amount of data, including measures of enzyme activity, selectivity, cell activity, in vivo pharmacology, solubility, permeability, metabolic stability and clearance to name just a few. Inability to normalize and standardize all that data for use by AI can pose challenges, not the least of which is cleaning up human errors contained in long-lived discovery and development effort.

Who identifies relevant data and defines the data to be utilized by AI can also be challenging, Kwak said. Lack of trust in AI, attributable to a lack of explicability in the results it generates, poses a third challenge.

Sang Ok Song, Standigm Inc.’s chief transformation officer, concurred, pointing out that a lack of high-quality, or clean and properly linked data as well as slow validation are all “major bottlenecks” in AI-supported drug discovery efforts. Song added that it was “a hard game to play,” due to the high barriers to entry in the AI health care industry and large amounts of investment required.

But challenges can extend beyond data issues, with AI specialists and domain knowledge experts providing a human element to the problem, too. Ineffectual communication, lack of trust, inability to find common goals or values due to profit-share agreements, different agendas and different operations sometimes hampers collaboration between the two groups.

Kwak cited Microsoft’s dispatch of AI specialists to Novartis AG, announced in October 2019, as an example of a successful collaboration. The Microsoft specialists that worked with Novartis domain knowledge experts in new drug development were treated more as Novartis employees than an external team, with rewards based on milestone achievements.

“A very well-defined and small question, finding robust, dependable data, building trust, starting with small successes, empathy, customizing the collaboration for the project, and engagement throughout the project” are all important, Kwak said.

Acknowledging that AI is as biased as the data it is fed, lowering often sky-high expectations, and a healthy dose of patience, it seems, could be important first steps to overcoming the hurdles to fully implementing AI in the drug discovery process.