AI-driven start-ups are key to evolution of drug discovery

The outlook for big pharma is challenging to say the least. It takes between 10 and 12 years to bring a molecule discovery to launch, while the average cost of the R&D process has almost doubled since 2010 to a whopping $2.168 billion per drug. Although companies are shelling out more and more cash to replenish their pipelines, they are simultaneously losing out on sales, with the expected return on investment from drug development dropping from 10.1% in 2010 to 1.9% in 2018. These figures are alarming at best and reflect an industry in crisis.

AI has become a serious contender for remedying the life sciences community’s biggest problems: efficiency and cost-effectiveness. While there have been a slew of start-ups leveraging AI-enabled solutions to tackle various issues across the pharmaceutical value chain, one area stands out for its game-changing potential: AI in drug discovery. This process currently accounts for one-third of the cost of development, but AI can help rein in this spending, as well as speed up the research process and improve outcomes. Ultimately this would enable far more competitive R&D strategies. Anyone who has a vested interest in the pharmaceutical industry is thus keeping a keen eye on this space.

Analysing the state of play gives rise to a complex dynamic. AI-driven start-ups focused on drug discovery are highly visible in this market, but universities, pharma giants and tech titans are also investing in these technologies. Analysis of patent data shows that it is the industry’s most established players (eg, GSK and Roche) that possess the most robust portfolios; yet the small start-ups enjoy significant advantages in terms of asset quality and technical expertise. As things continue to progress, it will be these smaller, specialised businesses that could drive much of the market.

Evolution of AI in drug discovery

The pharmaceutical industry has been slow on the uptake of AI, and even now it has yet to go all in. A 2018 survey by Pistoia Alliance found that 72% of life science professionals in Europe and the United States believe that the industry is lagging when it comes to AI development. Two unique barriers have prevented the widespread adoption of such technologies: the limited availability of clean and connected data, and the difficulty of hiring specialists who possess the necessary skill sets. These have been a drag on progress, but the market – pushed forward by AI start-ups – is now quickly gaining traction.

AI-driven start-ups focusing on drug discovery emerged on the scene just over 10 years ago, with a total of 20 active companies in 2009 (see Figure 1). These numbers picked up in earnest in 2015 when 31 new start-ups were formed, with an additional 36 entering the market in 2017. While the level of start-up activity has certainly fallen since then, the latest figures reveal that there are well over 200 businesses focused on leveraging AI in drug discovery. These companies have received just under $5.2 billion in funding from venture capitalists, with $1.4 billion invested in 2018 alone.

Tracking the proliferation of AI-driven start-ups is one way of understanding the evolution of the drug discovery landscape. Another is to look at the deal-making activity between large pharmaceutical companies and these smaller businesses (see Figure 2). There has been a surge in collaborations – from research deals to partnerships – between start-ups and pharma giants since 2017. Several big-ticket deals have also been signed, including Atomwise and Hanso Pharma’s collaboration, which was announced last year and is worth up to $1.5 billion, as well as Exscientia and Celgene’s $25 million three-year partnership.

When it comes to patents, traditional players take the lead

Given the disruptive nature of AI-driven start-ups and their high visibility, they are often a key focus when gauging development in the market. However, they are far from alone in pursuing AI solutions for drug discovery. C-suite executives have an increasing appetite for these capabilities and every major pharmaceutical player has a strategy in place to tap into these solutions. One cannot forget the tech giants such as Alphabet and Microsoft either. These heavyweights are already AI savvy and are moving fast into the healthcare sector.

Sumair Riyaz and S Rajan Kumar, who lead business development and technical analysis at Dolcera, drilled down into research and patent data to obtain a more nuanced view of this high-growth market. A search for peer-reviewed papers published on the use of AI in drug development shows a significant uptick in interest in the field, with 1,079 papers published in 2019, nearly three times the number published in 2015 (see Figure 3).

Published patents relating to AI in drug discovery follow a similar trend, although it is worth noting that activity levels were relatively high as far back as 2006. Since 2008 there has been steady growth in these filings, with a distinct jump in 2019 (see Figure 4).

Figure 4. Patent publication year – trends

Major pharmaceutical giants have taken a multi-pronged approach to integrating AI into the drug discovery process. This includes investing in relevant start-ups, creating collaborations with leading players, tech giants and research centres, and shoring up their own in-house teams by hiring AI experts and data analysts. While the latter is particularly challenging, some of the industry’s heavyweights have made significant headway here. Novartis, for example, made it its mission in 2018 to focus on powering itself with data science and digital technologies, and began restructuring its Global Drug Development IT infrastructure to achieve this.

Rival GlaxoSmithKline (GSK) has moved along the same lines, identifying domain expertise, data and deep learning as the key tools for improving drug discovery. It was one of the first off the mark, hiring Samsung’s chief data officer Mark Ramsey in 2015. Before he left in 2019, Ramsey was responsible for building a platform that uses large-scale data analytics to speed up – and improve – the drug development process. GSK is one of the most active implementers of emerging technologies; as of July 2019 the company’s AI team numbered about 50, with the goal of hiring an additional 80 specialists in 2020, according to The Guardian.

It is hardly surprising then that GSK is the leading pharmaceutical company in terms of published patent applications relating to AI in drug discovery (see Figure 5). With 4,463 published patents, it dwarfs the portfolios of Roche and Johnson & Johnson in second and third place, which own 1,876 and 941 published patents, respectively. Dolcera conducted additional analysis to understand which companies owned the strongest portfolios, using its patent ranking system and selecting patents with a ranking of three and over (the scale runs from zero to five) to show only high-quality assets. GSK yet again reigns supreme with 2,680 patents ranking three and above, followed by Roche, Johnson & Johnson, AstraZeneca, Novartis and Pfizer, which own 1,041, 602, 559, 403 and 399 high-strength patents, respectively.

Academic institutions do not take up much of the limelight in the industry-wide discussion about AI drug discovery solutions. However, universities are highly innovative and often have access to data that is critical to developing AI tools, making them valuable market partners. Further, they often house and nurture research teams that go on to become start-ups. One of the latest entrants to the market, Genesis Therapeutics, was actually spun out of Stanford University following research conducted by its principal founder. Insilico Medicine, which secured $37 million in Series B funding in September 2019, was in an incubator programme through a joint initiative between Johns Hopkins and the city of Baltimore – it is now based in Johns Hopkins University’s Emerging Technology Centres.

The regents of the University of California own the largest portfolio among academic institutes with 1,258 published patents (see Figure 6). Trailing behind them are Harvard University, with 726 published patents, and the board of regents of the University of Texas System and Massachusetts Institute of Technology (MIT), both with 692 published patents. Notably, each of the top academic institutes featured in this list are US-based.

However, this ranking is shaken up when patent strength is isolated. The regents of the University of California retain the number one spot, with 578 assets ranked three and above, but Stanford edges out the University of Texas System to take second place with a portfolio of 391 published patents. It is a close race for third, with the University of Texas System coming in with 371 high-quality assets and Harvard University with 369.

Various tech giants have also emerged as significant players. They were some of the first to foray into the implementation of AI, often through inorganic growth, and have taken a keen interest in gaining a stake in the healthcare sector. Yet, their patent holdings are no match for those of either the pharmaceutical giants or the top universities (see Figure 7).

Alphabet owns the largest portfolio, with 571 assets, followed by IBM (386) and Microsoft (260). IBM’s AI platform, Watson, offers support for businesses in the healthcare sector (among others). However, there have been reports in Chemical & Engineering News and STAT that IBM will no longer be offering Watson for drug discovery; it will support those currently using the platform but its focus within Watson Health will shift to clinical development. Although Microsoft owns a relatively smaller holding, the company has already signed collaboration deals with AstraZeneca and Novartis. Its alliance with Novartis, announced at the end of 2019, will focus on using AI to accelerate the company’s drug discovery and development processes.

Significantly, the tech giant graph includes major Asian players Tencent, Huawei, Baidu and Alibaba. These companies have been proactive in the adoption of AI-enabled solutions. Tencent and Alibaba both have deals with XtalPi, a Boston-based business that was founded by a group of quantum physicists at MIT, and Tencent and Baidu have both struck up agreements with Atomwise, an AI-start-up that also calls Bayer and Merck partners.

The Asian heavyweights largely drop off the list when the focus is on patent quality, with only Huawei maintaining its ranking with five exceptional assets. Alphabet owns the largest portfolio of valuable patents, with 394 assets, followed by Microsoft (174) and IBM (173). Notably, Nvidia ranks alongside Huawei with five published patents of high quality.

Curiously, when it comes to patent protection, AI-focused start-ups tend to have relatively small portfolios. BERG, Pharnext and DeepMind (which is owned by Alphabet) have the largest holdings (see Figure 8). Both BERG and Pharnext are some of the older start-ups in the market, founded in 2009 and 2007, which may explain the higher volume of assets. As for DeepMind, it has likely benefited from having such a patent-savvy parent company.

Although the AI firms have much smaller portfolios, a greater proportion of their assets are of higher quality (see Figure 9). Of BERG’s 170 patents relating to drug discovery, 126 receive a ranking of three and above. Similarly, 116 of Pharnext’s assets receive high scores for strength.

Future of the industry: start-ups will be calling the shots – for now

The patent data sheds light on the state of play, with a number of key takeaways. The first is that the tech giants own fairly sizable portfolios. Second, the major pharmaceutical players are all investing in AI technologies in a bid to accelerate the drug discovery process; as part of this they are hiring specialists to strengthen their in-house capabilities. This is the source of their patents and they are probably hoping to tap into these assets in the long run. Finally, not a lot of the AI-driven start-ups have patents, but those that do have relatively high-quality portfolios. As always though, patents are not everything and a significant amount of these companies’ strength is based on their technical expertise – although their IP assets suggest that they may become attractive acquisition targets for the largest pharma companies or even an ambitious tech giant.

While we have been able to extract these insights from the data, it is worth noting that it is still difficult to get an in-depth understanding of AI in drug discovery from a patent perspective because it is a nascent industry. Additional analysis, conducted by Randolph Square, a patent-centric analytics and litigation funding boutique, showed just how much this was the case.

Randolph Square’s analytics process focuses on constructing a “synthetic cohort” using a variety of semantic and Cooperative Patent Classification-based methods. Its process found AI in drug discovery to be a fragmented, immature innovation space lacking clear winners. Further, there was a high degree of inventor-owned assets – a characteristic of industries that have yet to consolidate through extensive M&A. Another noteworthy finding was an anomalously high degree of semantic similarity among competitors’ patents, begging the question of whether we are looking at the early pioneers in an emergent area of technological significance or the latest lexicon of what amounts to a bubble in the demand for venture capital.

Randolph Square’s analysis touches on something important. AI-enabled solutions are expected to have disruptive potential, but there are still a lot of questions surrounding the viability of these programmes and how effective they will prove to be. This is apparent in the venture capital funding, which hit $1.4 billion in 2018 but dropped to $1.1 billion in 2019. The fall in investment can be attributed to a number of causes, but one is that investors are waiting to see more tangible proof that these AI solutions work.

It is difficult to definitively assess the potential of AI-driven programmes, especially for biopharma companies that are evaluating young start-ups but that have only limited information on the effectiveness of their solutions. One thing that has been promising though – and that demonstrates the confidence that start-ups have in their own capabilities – is the fact that, so far, many have not followed the typical route of being acquired by a larger entity. Instead, they have actively sought partnerships and collaborations with the industry’s major players. Further, the market was handed a strong win in January this year when the first drug designed entirely using AI entered human clinical trials. It was a compound created by Exscientia and reached trial in less than a year – five times faster than usual.

There are plenty of challenges for big pharma to build up their own capabilities, ranging from cost to scarcity of scientific or engineering talent, which is why there has been so much deal making between these companies and the leading AI start-ups. These relationships offer some clear benefits, providing the pharma company with an AI solution that is tailored to its specific data and handing the AI-focused company an opportunity to improve its technological capabilities through greater access to data.

It may be that AI-driven start-ups become highly sophisticated and end up in possession of such high-quality patents that they begin to be bought outright. But their current strategy of remaining independent, if profitable now, could be even more so in the future. Should their potential be realised, they could be in line for a healthy royalty stream and sub-licensing on the sale of drugs created through collaboration, as well as cashing in on charging for access to their technology. It looks as if start-ups know exactly what their value is to the market and that they will be the players to watch as AI unfolds in drug discovery.