A $20bn industry by 2024?
AI for drug discovery will be driven by biopharma and the rise of Asian tigers, says Insilico CEO
Artificial intelligence is among one of the latest buzzwords in the industry. Yet, the CEO of Insilico Medicine, a global biotech using AI for drug discovery and aging research, noted the hype has been mostly in the press, not the bank.
“It is not the dot-com boom,” said Alex Zhavoronkov, PhD. “Big pharma is a very, very conservative industry.”
Over five years, Insilico Medicine has raised approximately $15m – compared to the IT sector, “it is night and day,” Zhavoronkov said.
An example of this, Google (now Alphabet) in 2014 acquired DeepMind for more than half a billion dollars. The sale of the company, at the time only four-years-old with a few employees, was a “historic moment” in artificial intelligence (AI), Zhavoronkov told us, one which “triggered the hype wave but played out very nicely for Google.”
Today, DeepMind employs more than 850 people, fueled by credible PR stunts demonstrating how AI can outperform human experts. Zhavoronkov said the marketing impacts were comparable to the Olympics.
“This does not happen in pharma. Instead of partnering and acquiring, pharma companies spend enormous resources hiring the talent and building their own expertise. The pilots they propose are very old-school and do little good for them internally or to the external partners,” he said. “And when they partner, it is with the companies run by the ex-pharma people, not so much AI.”
What will happen?
Zhavoronkov predicts this year will see more AI companies raising funds to focus on their own development efforts, not wanting to participate in pilot studies with pharma, “as these pilots take years and rarely result in the deals that justify high valuations required by the investors.”
According to Zhavoronkov, the only company which showed that it is possible to make substantial deals with big pharma is Exscientia, which recently announced deals with Sanofi, GSK, Roche, and Celgene.
“While the machine learning component there is likely to be a bit older school, Exscientia is a true leader when it comes to making pharma deals. They manage to explain the cost savings and do the work more efficiently than the internal teams,” he explained.
“Technology leadership is not as important in pharma as the big names associated with the company," Zhavoronkov added. "Some venture capitalists recognize this trend and are willing to bet big. They look at the many models and then build their own companies."
When asked about Insitro, a very well-funded company with investors including Foresite Capital and others, Zhavoronkov said he expects that they should be successful. He explained, “They brought several smart people together and invested a substantial amount. They are likely to announce a couple [of] big pharma deals to show external validation and will raise an enormous amount of funding to essentially become a pharma company.”
As per DeepMind and Google Health, Zhavoronkov expects the companies to become more serious about drug discovery, citing the buzz created from AlphaFold, a system which uses genomic data to predict protein structure and create 3D models.
While the result may not be revolutionary, according to Zhavoronkov, if structure prediction and molecular generation are brought together, the AlphaFold system would certainly outperform a medicinal chemist in a “Go game” exercise. This exercise, demonstrated by DeepMind’s AlphaGo, is said to be one of the most challenging games for AI.
In line with this, combining deep reinforcement learning with generative adversarial networks (GANs) – machine learning systems in which two neural networks compete in a zero-sum game framework – and domain expertise for small molecule design is going to be a trend in 2019, Zhavoronkov said.
Insilico first presented GAN-generated molecules in 2016 and while people were skeptical, the technology is now being embraced and democratized, as recently exemplified by the company’s launch of Pandomics.com. The multi-omics target identification and drug scoring platform provides open access to several of the company’s internal tools.
This ability to target proteins is one of three key areas in which AI will make most of the difference in pharma R&D, said Zhavoronkov.
The second is a “massive acceleration” in the time it takes to generate novel molecules in order to validate these targets in disease-relevant models. “I think that it is already possible to go from a protein target with a structure to an IND-ready molecule in less than a year and with the giants like WuXi AppTec, which is the ‘Ferrari of synthetic chemistry’ it is possible to synthesize the molecules very, very quickly,” he said.
Additionally, AI will be important for knowledge discovery and hypothesis generation, with knowledge graphs now commonplace. This technology is important “for augmenting the target identification process and small molecule generation to increase the confidence in the target and the molecule,” Zhavoronkov explained.
The combination of these three areas will help significantly accelerate preclinical R&D, he said – and is likely to this year see the first molecules designed entirely using AI submitting an investigational new drug (IND) application.
AI for drug discovery: A $20bn industry by 2024?
“Discovering new drugs using AI is one of the most challenging areas in biological sciences because of the high level of expertise required in both biopharmaceutical science and technology,” said Margaretta Colangelo, managing partner at the investment fund Deep Knowledge Ventures.
“To deliver tangible results teams must be capable of applying innovations in AI, machine learning, deep learning and big data analysis, and integrating them with precise quantitative metrics for digital medicine,” she told us.
In the next few years, Colangelo said the arrival of the first AI-designed drug on the market “will validate the entire AI for drug discovery industry, causing a massive increase of investments into the sector.”
Investment in AI for drug discovery startups increased from $200m in 2015 to more than $700m in 2018 – and the number of companies in this space increased by 20, according to a report published by Deep Knowledge Analytics, the analytical arm and subsidiary of Deep Knowledge Ventures.
“Demand for AI technologies and AI talent is growing in the pharma and health care industries and driving the formation of a new interdisciplinary field — data-driven drug discovery/health care,” added Colangelo.
The current leaders in this space currently are WuXi NextCODE, BenevolentAI, DeepMind Health, and Insilico Medicine, Colangelo said: “These companies are capable of hiring and retaining a high number of AI experts and biochemical specialists, have significant financing, have efficient business development strategies, have partnerships with pharma companies, and they are capable of building end to end solutions.”
Per the report, there were 350 investors in this space in Q1 2019 – 30 more than Q4 2018. Among the investment funds participating are Google Ventures, Tencent, Wuxi, Andreessen Horowitz, Khosla Ventures, and Sequoia Ventures.
In addition to strong financing, Colangelo said innovation is driven by highly skilled humans, cutting edge technology, and innovative business models.
“Biopharma companies capable of building strong AI divisions and acquiring the best AI startups will dominate the biopharma industry,” she said.
Though analysts have not reached a consensus as it pertains to the industry’s expected valuation, estimates range from $5bn to $20bn by 2024.
Bullish on China and betting on AI
Also in 2019, Zhavoronkov expects it will be the year of pharma in China. “I am very bullish on China,” he said, and for this reason, Insilico has R&D centers in Hong Kong and Taipei.
“With 1.4bn people and the government push for innovative medicines China will be the dominant force in the pharmaceutical industry. I only hope that the trade wars do not impact this important field,” he said. “Cancer, Alzheimer's and other diseases do not discriminate by the nation. Until there is a clear set of cures, a trade war in biotechnology R&D is pretty much a war on your own people.”
Zhavoronkov also predicts that the China-based pharmaceutical companies “pushed by the government to develop innovative medicines” will embrace AI-powered drug discovery “to leapfrog the years it takes to discover novel targets and generate great molecules.”
Smaller biopharmaceutical companies that do not have the burden of existing expensive legacy R&D departments will now be able to virtualize drug discovery and innovate faster and fail quickly.
And Insilico is betting on both of these trends: “Smaller biopharma companies unburdened by the internal legacy R&D and the rise of Asian tigers committed to innovative medicines and confident in the potential of AI.”
Big pharma, on the other hand, is unlikely to reach a competitive tipping point, Zhavoronkov said.
“Internal R&D is not something big pharma is good at,” he added. “I think that the only way big pharma can reach this tipping point is by making big bets on AI companies.”
So how does AI succeed in pharma?
Zhavoronkov thinks many AI firms and venture capitalists have begun to realize there are four approaches, which he described:
1. Develop your own molecules into drugs
2. Start a new AI company with an AI expert, raise enormous amounts of money, and make pharma pay a lot of money to partner with you
3. Present yourself as a very credible incremental player making one process more efficient and make deals at the board level
4. Work with the smaller biopharma companies or with the Asian companies that want to leapfrog the big pharma R&D processes; and if your technology works, they partner quickly and can provide access to experimental data
Insilico Medicine was recently named one of the world’s top AI drug development companies by Frost & Sullivan, CB Insights, NVIDIA, and Forbes. The company is focusing on target identification and compound generation using next-generation AI techniques. Nearly a quarter of Insilico’s research and development (R&D) budget is put toward AI for age-related diseases.