Via computer algorithms, artificial intelligence (AI) is capable of solving complex genomic data, such as the gene expression pattern of disease. On the one hand, AI is about to revolutionize many fields of the biopharmaceutical industry like drug design, drug screening and hit identification, clinical research and medical treatment. On the other hand, AI may also help lower drug prices, which marks another milestone in the history.

Although AI cannot completely eliminate the drawbacks of innovation process featuring inefficiency, time-intensiveness and trial-and-error, it can at least make a significant difference. This is the value that many AI experts emphasize. According to the statistics from the National Association of Pharmaceutical Research and Manufacturers (PhRMA), only about 12% of drugs entering clinical trials are finally approved by regulatory authorities. AI or machine learning, as well as more sophisticated extension technology "deep learning", aims to increase the success rate of drug development. Experts say that AI technology is likely to speed up research and more effectively turn candidate drugs into marketed drugs than traditional practices.

In one word, AI might usher in the industrialization of biotechnology and pharmaceutical innovation. Read below how industry experts view the impact exerted by AI.

Dr. Gini Deshpande, CEO of NuMedii, said that this does not mean that artificial intelligence will replace research scientists. He prefers the term “intelligence enhancement”, which combines human-computer intelligence to “simplify the discovery of related disease biology and thus shorten drug discovery time, but more importantly, significantly increase the likelihood of success."

Dr. Vijay Pande, general partner of Andreesen Horowtiz, pointed out that if AI is effective, it can have an impact on lowering drug prices in two ways. First, pharmaceutical companies do not have to pass on the cost of all clinical trial failures to payers. Matter-of-factly, 90% of drug candidates have never been approved. Second, by speeding up the time to market, companies will have more patent protection years to balance their research and development costs.

So far, artificial intelligence seems to be something far away from biotech and pharmaceutical companies, but they have begun to notice it recently. “It's the same as any new technology in the biosphere,” Dr. Pande said. “It takes a while for AI to be accepted. At some point AI will start to work, and that's where we work.”

The status quo of AI in drug development

The potential of AI for biomedical applications is exciting. In other areas, AI algorithms have trained machines to learn how to recognize faces, talk, drive cars, play games, and even compose music. As described in an article in Molecular Therapeutics: "The type of learning required for these tasks is representation learning; that is, after inputting raw data, the machine can detect or classify patterns or characterize information."

This makes artificial intelligence and its subsets, machine learning and deep learning well suited for mining large amounts of genotype and phenotypic data from global public and private databases, hospital and physician offices, academic research journals and personal wearable health monitoring equipment.

Dr. Pande of Andreessen Horowitz said: "I think an important thing that biologists have recognized in recent decades or even centuries is that biology is very complex. Biology is so complex that human brains can't understand all of this; our understanding of biological limitations has led to the complexity of late clinical trial results."

The proficiency of artificial intelligence in pattern recognition and feature learning and can more accurately predict outcomes, thus capable of revealing the complexity of the genetic level of the disease pathway, designing therapeutic interventions, and identifying patients who will benefit from it. AI has become the most promising research tool.

Dr. Pande said: "Biology may be too complicated for the human brain, but it may not be too complicated for certain types of AI; AI can integrate data in ways that are beyond the reach of the human brain, and then lead researchers to some interesting points."

So when can the strength of artificial intelligence algorithms be released? Most experts believe that 10 years is still needed before AI is fully integrated into the R&D department of biotechnology and pharmaceutical companies. But they also agree that this technology will become an important tool for drug development.

“I don’t know if there will be a place for biotech or pharmaceutical companies that don’t use AI in the future,” Mr. Smarason said. “This is a very comprehensive and universal technology, so I think if you don’t fully understand and accept AI technology, it will be very difficult for pharmaceutical companies to compete."

The initial attraction of AI to biopharmaceutical companies is to simplify R&D operations and increase the speed and success rate of clinical trials. PhRMA estimates that the average cost of developing a new drug is $2.6 billion. A report from TechEmergence examines AI applications across all industries, and the results show that AI can increase the success rate of new drug development from 12% to 14%, saving the biopharmaceutical industry billions of dollars.

The challenge of AI new drug development

However, most of these expected R&D savings are long-term effects, so here are still some challenges for AI to be applied in the biopharmaceutical industry.

Mr. Smarason said that no products that are directly related to AI have been approved. “But I will say that we will definitely see some important (drug) products related to AI coming out within the 10-year time frame.”

According to Dr. Hopkins, one of the challenges facing AI today is to ask “experts in the field of drug discovery to define reasonable questions. If the questions asked are too ambitious and the data is insufficient, the questions submitted to AI will be imperfect.”

Another challenge, as Dr. Deshpande said, is to get "high quality and consistent data to train algorithms. Currently data is usually kept in silos and spans multiple organizations."

Dr. Lanza added that overcoming the traditional R&D culture is also a challenge. He said: "Artificial intelligence means that it can't be explained in essence, but more as a 'black box.' I often hear that in order to make these predictions predict, scientists want to know how AI does it. This is usually the wrong way of thinking about AI. The key is that the signals in the data that these algorithms can see are too narrow or too wide for humans. Therefore, if we ask artificial intelligence to produce human-readable results, it may limit AI to solve the most interesting question."

Author's Bio: 

BOC Sciences