On average, the cost to develop a new drug is in the $2.6 billion range. According to the Eastern Research Group, it takes 10 to 15 years to develop a new medicine.
Most of this money is “wasted” in failed clinical trials. Eighty-six percent of the drugs will fail during clinical trials. Even if these drugs were promising in animal models, they did not meet the expectations to treat patients.
Developing new drugs quickly is not only good for the bottom line of pharmaceutical companies. It is also vital for patients, especially for those suffering from orphan diseases. It implies identifying a target on which the new chemical entities will interact to induce a positive outcome on the disease’s course.
Screening chemical libraries one after the next take a huge amount of time. Artificial Intelligence (AI) does not only involves self-driving cars or Science Fiction movies. Intelligent systems have the potential to reduce the cost and time needed to develop new medicines. Let’s check how AI systems can boost drug development in a post-Covid-19 world!
Drug Discovery & Drug Development 101
We recommend watching these two excellent videos to know more about the drug discovery and drug development processes. Those of you who want a quick introduction to this long and difficult process, please continue reading.
Developing a drug is a highly complex and standardized process. We will simplify it a bit and only present the major steps to bring the idea of a drug target to a new medicine. Remember that this process will take 10 years on average and costs $2.6 billion! No Star Trek warp speed for drug development!
The first two drug discovery steps (identification of a drug candidate and preclinical studies) will take an average of five to six years to complete. Out of the 10,000 molecules screened during step one, only ten will get into clinical trials. It is only one percent that will be tested in humans.
Identifying a drug candidate
The first step is to identify a molecule that will be active against the targetted disease. This is where Artificial Intelligence comes into play, but we will deal with this that later. It all starts with a target, a molecular target. In a few words, a drug target is an active site on which a molecule will bind.
For example, the widely used anti-cholesterol drug called statins will connect to a molecular site. It will inhibit the HMG-CoA (hydroxymethylglutaryl-coenzyme A) reductase enzyme involved in the synthesis of cholesterol. By blocking this biochemical reaction, cholesterol production will be limited, and the arteries will remain clean.
Once a target is identified, thousands of molecules will be tested via a High-Throughput screening process. It will start with a hit and a lead that will be chemically optimized for efficacy, pharmacokinetics, and pharmacodynamics.
Pre-clinical studies
Non-clinical studies or preclinical testings are the first steps before moving to humans trials. To get to this stage, an application will be submitted to the regulatory authorities: FDA in the USA, EMEA in Europe, and PMDA in Japan. We recommend that you contact our partner, HASHI Consulting, for all your drug development needs in Japan, but it is another story.
During the Investigational New Drug (IND) application process, Pharmacokinetics, Non-GLP Toxicology, GLP Toxicology, and Safety will be presented to support the dossier. At the same time, manufacturing will have to be set up using Good Manufacturing Practices (GMP).
Following the approval of the IND, it will be time to move to the clinical development phase. It is the last hurdle, but quite a huge one! Ninety-five percent of the new potential medicine will fail during this phase. Remember?
Clinical studies
Clinical studies are where the side-effects and real benefits of the new drug are assessed. Initially, only healthy volunteers will receive the product. Any serious toxicity will be a NO GO. During phase II, the efficacy is tested on a small group of patients.
If everything goes well, the number of patients enrolled will increase, and Phase III will begin. Depending on the end-point, the studies will take a couple of months or a couple of years. For example, it will be much quicker to assess a compound’s efficacy to treat migraine than a drug aimed at slowing down the evolution of neurodegenerative diseases.
All of the new medicines will undergo this process, both designed to ensure patient safety and clinical benefits. As a side note, we have all heard about Emergency Use Authorization (EUA) during the last months. A drug or diagnostic test used under EUA is still tested. The requirements are less drastic, though. Hence a faster approval timeline.
Differences between AI, Machines Learning, and Deep Learning
Talking about Artificial Intelligence, machine learning, and deep learning can be confusing. We all know it relies on some machine mimicking the human mind, but what are the real differences?
Alan Turing, the famous British Mathematician, is often credited as being the father of Artificial Intelligence. Hence, the Turing test was designed to test if a machine intelligence was indistinguishable from a human’s one. Artificial intelligence is basically dealing with correlations and extracting relationships invisible to human eyes.
Machine learning algorithms will be fed with structured or unstructured data. The ultimate goal is to identify hidden patterns and make predictions. For Deep Learning, computers will show a neural network to mimic how the brain works by increasing or decreasing the synaptic weight. Computer vision is the prototype of deep learning, and your self-driving car uses deep learning all the time to recognize differentiate a crossing grand-ma from the shadow of a tree.
When applied to drug discovery, Artificial Intelligence allows to:
- Reduce the time needed for drug discovery: The company Insilico Medicine discovered a new drug in 21 days and validated it in 25 days when it would normally take 2-3 years with standard methods. Thanks to Deep Learning, DDK1 kinase inhibitors were rapidly identified.
- Deliver accurate predictions on the accuracy and efficacy of new drugs.
- Broaden potential drug pipelines: By relying on data only and without any hint of apriori assumptions. Artificial Intelligence is perfect for detecting potential new therapeutic applications for drugs already in use.
Differences between Drug Repurposing and Drug Discovery
Drug repurposing is the principle of taking an existing drug and look for new pathologies where it could be useful. The benefits of drug repurposing are clear. Using the base of an already validated drug to treat a different disease will save time and ensure better toxicity profiles.
Using Artificial Intelligence in Drug Discovery
Drug discovery is very similar to being a locksmith. What is needed is to identify a key (molecules) that will fit perfectly with the binding site’s shape and allows the therapeutic benefits to occur. One of the main hurdles in drug discovery lies in the lack of precise knowledge of both the key and the lock’s tridimensional structures.
It is where computers shine. From number crunching glorified calculators dealing with only specific tasks a couple of years ago, they now can learn, run models, solve problems and make decisions.
Artificial Intelligence has the potential to enhance all of the steps of the drug discovery process.
- AI-Enhanced Identification: Instead of validating targets based on genomic and proteomic data and screening huge chemical libraries, AI can rationalize the target identification,
- AI-Enhanced Engineering: Most of the selected targets for drugs are proteins. Hence the need for precise tridimensional structure analyses. To get back to our analogy of the locksmith. Artificial intelligence can find locks that will perfectly fit the existing key and even to modify or build new keys,
- AI-Virtual Screening: Instead of spending a huge amount of time testing the compounds in vitro, AI can save a lot of time by conducting virtual screenings,
- AI-Enhanced Optimization: Chemical entities are rarely perfect. They need to be optimized to be able to reach their target with maximum efficiency. Here again, AI can help in the development of best-suited leads,
- AI-Enhanced CMC: By building virtual chemical factories, AI is used during the Control Manufacturing and Control phases. The precise selection of patients to be part of the clinical phases could also dramatically enhance the outcomes. By maximizing the therapeutic effects, costs decrease,
- AI-Enhanced Monitoring: Following the release of a new medicine on the market, the pharmacovigilance phase will begin. By linking adverse effects negatively, AI can also improve compliance and drugs’ safe use.
Using AI against Covid-19
“The most fruitful basis for discovering a new drug is to start with an old drug.”
Sir JAMES BLACK, Nobel Prize-winner (1924-2010)
Drugs with already known safety profiles have the potential to be brought much quicker to the patients. The use of Remdesevir is probably the most well know for Covid-19. Using Knowledge graphs containing the relationships between different kinds of medical data (drug, patient, efficacy, chemistry), the company benevolent AI predicted that a drug to treat rheumatoid arthritis could be useful to great Covid-19. More than 39 types of relationships, including 24 million scientific articles, were processed on Amazon Web Service to identify 41 potential repurposable drugs.
Using Deep Learning methods and neural networks, scientists found several known antiviral drugs such as Atazavanir, Remdesevir, Efavirenz, Ritonavir, and Dolutegravir had potential in treating SARS-CoV-2 infections.
By combining Physicians, Computer scientists, pharmacologists, biologists, clinical and scientific data, Artificial Intelligence proved useful when fighting Covid-19. Challenges remain due to the unequal quality of the information. Big Pharmaceutical companies invest billions in developing new drugs and are reluctant to disclose patented information that may prove very useful for finding new indications for already marketed drugs.
Best private companies using AI in drug discovery
This section will present what we think are the best of the best when using AI for drug discovery. All of the Big Pharmaceuticals companies already use of Artificial Intelligence. Pfizer is famous for partnering with IBM and the famous Watson Health.
We will not focus on these big players. On the contrary, we will highlight the startups that may well change the future of drug discovery and development using revolutionary AI technologies. This list is far from being exhaustive, of course, and we apologize to the amazing companies that we did not highlight.
According to Crunchbase, the top Ai drug-discovery companies raised more than 2 billion dollars in the past 10 years to support their growth. More than 250 companies work at the intersection of biology and Artificial Intelligence. Our selection is based on following smart-money flow and how big drug makers rushed to conclude deals with these Startups.
It is not frequent for a company only 2 years old to have raised more than $243 million. Daphne Koller, the founder of the company, is certainly not new to machine learning. She co-founded Coursera with Andrew Ng, who is often considered the pope of Artificial Intelligence. Baidu and Google will certainly agree with this comparison.
Insitro is at the intersection of Bio-Pharma and machine learning. The company relies on human data sets containing molecular and clinical data, used to train computer models. Pluripotent cells with the ability to differentiate in any cells are modified to model diseases.
Machine learning will track the subtle differences between the sick and healthy cells to revert the cells to their healthy state. Combining biology and Artificial Intelligence in a single platform is mind-blowing, to say the least. Ten years ago, I developed High Content Screening models to screen for new drugs on cell cultures. We only tested a couple of dozens of compounds daily. Using machine learning to speed up the process, increase the accuracy, and find non-trivial correlation is a huge step forward.
After Gilead for liver diseases, the company recently announced a new deal with Bristol Myers Squibb to develop new drugs aimed at amyotrophic lateral sclerosis (ALS) and frontotemporal dementia (FTD). The potential of this deal is in the $2 billion range.
Since its inception in 2013, the London based company raised $292 million and has already concluded a major partnership with AstraZeneca for the development of new medicines in Idiopathic Pulmonary Fibrosis (IPF) and Chronic Kidney Disease (CKD). It is not the first one for this drug-development unicorn. They partner with Novartis to speed up the Pharmaceutical Giant’s oncology pipeline development.
BenevolentAI is data-hungry. By analyzing huge amounts of data from clinical trial reports up to scientific papers, BenevolentAI aims at identifying new relationships. The hypotheses found by the system are not always right, but they are unbiased and have the potential to unveil unknown correlations.
The power of the AI engine developed by BenevolentAI was recently in the spotlight for identifying the potential benefits of using a rheumatoid arthritis drug, Baricitinib, to treat Covid-19. When used in combination with Remdesevir, the drug combination significantly fasten the recovery of the patients.
The success of Baricitinib is a perfect example of the benefits of AI in drug repurposing. A drug candidate was identified in less than 6 months, and the preliminary clinical studies were successfully conducted. This is unprecedented and paves the way for a completely new approach to drug discovery.
In the last six years, Insilico Medicine raised $51.3 million. Even if it appears to be less than other companies, the Hong Kong-based company has attracted major drug makers’ and investors’ attention.
After signing a partnership with Pfizer and Boehringer-Ingelheim earlier this year, the company announced another deal with the Chinese drugmaker Jiangsu Chia Tai Fenghai Pharmaceutical. The last $37 million round of investment attracted very prestigious VC in Asia and the US such as Qiming Venture Partners, Eight Roads, F-Prime Capital, Lilly Asia Ventures, Sinovation Ventures, Baidu Ventures, Pavilion Capital, BOLD Capital Partners.
Even if Insilico Medicine uses its Artificial Intelligence platform to focus on cancer and age-related diseases, they recently entered a major strategic agreement with British company Arctoris in the discovery of Covid-19 drugs.
Arctoris is the world’s first fully automated drug discovery platform. They develop chemical or cell-based high throughput assays. Combining AI target and drug identification with the power of automated screening platforms is an amazing way to speed up the discovery phase and the validation.
Since its IPO on July 16th, the share price of Relay Therapeutics (Nasdaq: RLAY) has more than double. Founded in 2015 in Cambridge, Mass, the company raised more than $520 million. An IPO was the next stage for this company’s development. They specialize in using AI to model how proteins move inside cells.
By analyzing the movements, shapes, and interactions, the company aimed mostly at the oncology market. The basic concept is to leverage AI computation’s power and AWS platform to identify small molecules that will ultimately modulate proteins’ functions.
As described by Krishna Yeshwant, Managing Partner of GV Fund and one of the early investors in the Relay therapeutics: “From my perspective as a physician and a computer scientist, Relay’s approach to drug discovery signifies the best computational and medical advancements of the past century.”
In Summary
Using Artificial Intelligence in drug discovery & development can bring life-saving drugs to patients suffering from orphan diseases. It somehow involves a paradigm shift in the way big pharmaceutical companies apprehend drug discovery.
Drugmakers will have to rely more on external partners such as technology companies. New partnerships will form. Google DeepMind AI platform is part of the Sanofi drug discovery strategy. Strong AI development is not the only bottleneck. People from different backgrounds will need to share knowledge and find a common language.
Building multi-disciplinary teams is the key to success. It will take time for the glass ceiling to break and for Artificial Intelligence to reach its real potential. The Covid-19 outbreak was the first live test of the amazing potential of combining technologies and expertise. We hope that computer scientists and biologists will work hand in hand in developing drugs that will ultimately benefit the patients and society in the years to come.
