Sumitomo Dainippon Pharma Co., Ltd, Japan and Exscientia Ltd., UK have announced the Phase I clinical trial of DSP-1181, a drug created using Artificial Intelligence (AI). The drug aims to treat obsessive-compulsive disorder (OCD). The trial begins in Japan from March 2020.

The company claims it is the first instance of using AI to develop a drug that’s being tested on humans.

Pharmaceutical expertise was contributed by Sumitomo Dainippon Pharma. Exscientia, on the other hand, provided the technology by using its Centaur Chemist AI platform.

In another instance, Deep Genomics, a Canadian company, had announced that it had used AI to fully understand Wilson’s disease. It has also used AI to detect a potential treatment for the same. Wilson’s disease is a rare genetic disorder where excess copper builds up in the patient’s body, often reaching life-threatening levels.

Is computing technology beginning to replace human researchers?

Will these developments set the standards for research in other fields?

Are we on the cusp of a new scientific revolution?

The contribution of modern technology in pharmaceutical industry is all too well known.

In this post, we begin by looking at how Exscientia used AI for drug discovery. We discuss whether AI is indispensable in the pharmaceutical industry.

Next, we look at a few pharma companies using AI. We also ask if AI in drug discovery is over-hyped. Finally, we look at the key challenges for wide-scale AI adoption in pharma companies.

How AI platform is used for drug development

To understand how AI develops drugs, let us understand the standard cycle of development of drugs.

Researchers identify a target protein that’s causing the disease. They study such proteins carefully and for a long time. Otherwise, there’s a big risk of losing huge amount of money on the wrong protein. Also, there’s an added risk that the protein would be related to the disease, but isn’t the one that’s causing the disease.

Next, the research process tries to find a compound or a molecule that would influence the protein. In order to influence the disease-causing protein correctly, the compound should be able to alter the protein. Due to this alteration, the protein will no longer be able to continue contributing to the disease.

During this process, inefficient compounds are tossed aside and only safe, efficient compounds are taken further.

So what is the role of AI in drug discovery and development?

Because there are hundreds and thousands of molecules out there, human researchers cannot manually test each of these molecules.

Yet, without testing each of them, there’d be no way of knowing which molecule would be the most appropriate to fight a certain disease.

So this is what AI platforms do. First, experts will feed in them parameters. They rummage through all the molecules. Each of these molecules is compared against the parameters.

Because it’s an intelligent system, the AI platform will keep learning and thereby identify one or more compounds that it finds most equipped to fight the disease.

How data is fed into AI for drug development

Today, research, feedback, reports, patient records and a whole lot of other things add massive amount of data on each disease. It is becoming close to impossible for humans to process or utilize all that data. Artificial Intelligence systems, on the other hand, are perfectly equipped to sift through all the data and make meaningful interpretations out of that.

There many, many channels of feeding data to the AI system for drug discovery and development.

One source of the data is, obviously, patients suffering from that data. This data is collected from patients at different stages of the disease.

But there’s more.

Data is also collected from people without the disease. Deep-learning programs run both the types of data and learn more about proteins whose presence makes a difference between a healthy patient and an ill one.

The machine learning abilities of the system strives to find and establish connections between proteins and diseases.

The importance of AI in drug development

As mentioned earlier, the huge amount of data that we produce isn’t easy to handle for humans. Here are some reasons why AI is becoming more important to the pharmaceutical industry:

  • Costs: The cost of bringing drugs to market has roughly doubled in the decade 2003 to 2013. Also, the returns on research have dropped from 10% to less than 2%. AI, with its accuracy, has the promise of improving this.
  • Speed: The lab-to-market time has increased to 12 years. If AI can really deliver as some people hope today, regulatory agencies could be more trusting. That means AI-developed drugs could be given a pass over animal testing models and move straight to patients.
  • Innovation: It might sound like a bit of exaggeration, but drugs for simple diseases have all been discovered. The ones that haven’t found any cure are the ones that are complex. Drugs for such diseases are difficult too, and AI with its deep-learning mode might turn out to be the right solution.
  • Bias: Human researchers, no matter how hard they try, might often be limited by their personal preferences and biases. As a result, they may chase compounds and proteins based on their bias and hunches. Such an approach costs huge amount of money. AI can be free from such prejudice, making the process more cost-effective and less error-prone.

Which pharma companies are using AI to develop drugs

Here are the major companies that are using AI to develop drugs:

  • Genetech is looking for cancer treatment with the help of the AI system of GNS Healthcare.
  • Sanofi is working with Exscientia on metabolic-disease therapies.
  • Atomwise is trying to find new treatment routes for drugs that are already found and in use. It is interesting to note that the technology used here is the one that’s used in facial recognition.
  • Deep Genomics has already announced that it has understood Wilson’s disease, with the aid of Artificial Intelligence.
  • Lantern Pharma is using AI to sift through the records of failed drug trials in order to apply corrections.
  • Pfizer is using IBM Watson. The objective is find cancer drugs, or more specifically immuno-oncology drugs.

What can AI do in future for the pharmaceutical industry

Despite all the claims by various experts (both in pharma and AI sector), there are many who think much of this is over-hyped.

Nevertheless, AI holds a lot of promise. Here are some of the expectations on what AI can do for diagnosis, drug development and treatment in future:

  • Deep learning could create and make meaning out of a large pool of annonymized data collected from all over the world.
  • Artificial Intelligence may aid in early detection of dangerous diseases like cancer.
  • AI would be able to find individual compounds that can act on just the right proteins, without impacting or disturbing the rest.
  • AI would ultimately be competent enough to knock off a few years from the drug-development cycle. In other words, drugs would hit markets and benefit patients earlier.
  • Sophisticated and specially trained AI systems may be able to provide patient prognosis.
  • Once regulators and researchers can trust AI enough, we may see many drugs permitted to pass over animal testing and directly moving to human trials.

Challenges for AI in drug development

While the future expectations mentioned above sound exciting, there are a sizeable number of challenges that AI will have to battle before machine learning, deep learning and artificial intelligence can significantly contribute to drug development.

Here are the top 5 challenges AI faces in the pharma sector:

Challenge 1: Absence of clear regulations. To be fair, this isn’t a case of regulators going slow. The fact is there aren’t enough precedents – at least, not yet – for regulators to form appropriate and encouraging laws. And let’s not forget the goals of regulations and innovations are often contradictory. The first is always slow in accepting in change while the second is in a hurry to usher change.

Challenge 2: Poor quality of data. In spite of tall claims, the fact remains that bad data will only produce bad results. While we do have a lot of data today, we do know a lot of is bad. By bad data, we mean unreliable, inconsistent or simply poorly structured data. In all these cases, AI will likely come up with poor solutions.

Challenge 3: Quantity of data. Do we really have a lot of data? Not always. As a matter of fact, there are a huge number of diseases where data is conspicuous by its near-absence. Because only rich data can produce results – at least as of now – we need AI systems that can make sense out of small data. In contrast, social credit systems in China had humungous data to learn from.

Challenge 4: Lack of trust. How many patients might be willing to trust drugs developed by AI? Because the mechanism is far from being in place, the acceptance of such drugs will take time.

Challenge 5: Misdirected research. There is always a good chance that AI will expend all its energy only to come up with drugs that are already discovered. While this challenge can be handled with relative ease as compared to the other challenges, a company always runs the risk of losing a lot of money this way.

Concluding remarks

Just like any other nascent technology, a lot of mystery, fascination and distrust surround the use of artificial intelligence in development of medicines.

Yes, we are beginning to see the first signs of momentous tremendous changes AI might bring in. However, the AI technology itself isn’t advanced enough to fully understand and independently design less complicated machines. In that situation, the skepticism of critics is understandable.

The fact remains that we will need more information, more studies, more endorsements before we can fully accept AI as a reliable drug development tool. Till then, we will have to continue double-checking everything.


1. Mayo Clinic

2. Scientific American

3. Nature

4. Wired