MedCity Influencers, Pharma

This is how AI will shape drug discovery

AI software is expanding the universe of screenable compounds and will deliver higher hit rates as compounds move through the pharma value chain.

As healthcare venture capitalists, we specialize in technologies that have the potential to transform large markets in healthcare. Earlier this year, we predicted that 2018 would be a year of venture and M&A activity across the pharma value chain due to where we are in the business cycle, a permissive FDA and recent pro-business tax reform. CNBC reported on an upcoming wave of deals as pharma companies put their repatriated cash balances to work in M&A. As VCs investing ahead of this wave, we asked ourselves where transformative technologies, like artificial intelligence and machine learning, might generate substantial pharma value creation as we’ve observed in other industries. We believe that AI will change the pharma R&D industry.

Technology VC in pharma R&D

An oft-quoted statistic is that it takes 12 years and almost $2 billion dollars to bring an average drug to market. That is the raison d’être for the biotech VC industry, which saw $10 billion of capital invested in startups in 2017 alone, a 50 percent increase on 2016 according to Pitchbook. In the world of technology (vs. molecule) investing, much attention has focused on using technology to augment downstream drug development and commercialization, such as making clinical trials more effective. This has given rise to large acquisitions like the purchase of Quintiles by IMS Health, now known as IQVIA. The trend has also laid the groundwork for the success of high growth startups like Evidation Health, (a portfolio company with our firm) which virtualizes clinical trials by capturing digital biomarkers and behavioral data, and Science37, which has developed a new contract research organization model to manage trials with a geographically dispersed patient population.

By using software to increase reach, speed and accuracy, these startups move biopharma assets through the phases of clinical trials more expediently, and participate in the associated economics of bringing a drug to market.

Upstream opportunity for AI

In 2018, we expect a surge of venture activity farther upstream, in the pre-clinical trials segment of the value chain referred to in pharma parlance as the research in R&D. This is where new drug targets are being identified and new drug compounds are being evaluated against those targets for characteristics such as efficacy, potency, and toxicity. Drug discovery is the first step in the process of identifying new medicines. It has two major parts: identification of a drug target (the biological cause of a disease) and identification of compounds that may have therapeutic benefits for the drug target (the chemistry that will treat it).

In traditional models, compound identification starts with 10,000 to 15,000 compounds, which are whittled down to approximately 250 compounds at the pre-clinical stage, giving rise to just five compounds entering the first phase of clinical trials, and a single compound at the very end of the process. Today, incumbent processes are centered around a laborious process called high throughput screening, which uses robotic automation to assay the biological activity of a few hundred thousand drug-like compounds in the hunt for a molecule that’s effective against a specific drug target in vitro (or in a lab). That sounds like a lot until you realize that most biopharma companies hold compound libraries that number in the millions and third-party vendors offer access to more than half a billion compounds.

Land grab for R&D data

This entire process is being redefined by artificial intelligence, by using AI models to screen orders of magnitude more compounds in silico (i.e., through computer simulation). Several software companies that have emerged were early leaders in this space by lining up major partnerships with large pharma companies while raising double-digit millions of venture capital dollars. Drug discovery is an enormous market that can sustain multiple large companies working on different targets and different compounds. However, our expectation is that companies with access to the largest and richest datasets across private, public and academic sources – not just for the compounds that previously succeeded, but the ones that also failed during the process – will ultimately win. We are entering an era marked by a land grab for R&D data.

Leading startups are separating themselves from a growing field. Atomwise applies AI to combined private, public and academic datasets to identify new drug candidates. They have formed commercial and research partnerships with dozens of large biopharma companies including AbbVie and Merck, agrochemical companies like Monsanto and academic institutions via a translational research partnership known as AIMS. Recursion Pharmaceuticals, which has raised $80 million in venture capital funding, has a drug discovery platform designed to interrogate complex biological interactions. BenevolentAI has raised almost $160 million to mine trials and academic data of successful and failed drugs to identify new potential therapy purposes – the company’s last valuation exceeded $1.7 billion.

Other companies in this space that have raised millions in VC funding include:

  •  NuMedii which identifies new disease targets for existing molecules;
  • Verge Genomics which identifies genetic biomarkers related to brain disease and potential drugs that impact them;
  • Numerate which has a drug engineering platform focused on small molecule drug design; and
  • TwoXar, which also focuses on drug repurposing.

What does success look like?

One way to size the opportunity of applying software to enhance each stage of pharma R&D is by breaking down the cost-per-drug launch at each stage of the process, which is well documented in a study conducted by Paul et al in 2010. This research showed that on a cost-per-drug-launch basis more money is spent upstream in target identification, hit discovery and lead optimization than downstream in each phase of clinical trials. On average $700 million per drug launch is spent upstream before a compound even enters the clinic, due to the sheer number of compounds being tested at the earlier stages. It is a numbers game, and this is where software is a game changer over established practices.

AI software companies that successfully identify these compounds can build an extremely valuable portfolio of pre-clinical assets, whether they choose to do that independently (as a biotech company), or via licensing and royalty-based arrangements with their biopharma customers (as a software platform). There is an avid pharma acquisition market for early stage companies that have identified valuable compounds but have yet to see any through final FDA approval.

3 predictions for AI in pharma R&D in the year ahead

Prediction 1: Drug discovery is being redefined by advances in technology. AI and machine learning software is expanding the universe of screenable compounds and will deliver higher hit rates as compounds move through the pharma value chain. The race is on for data.

Prediction 2: Within the next 10 years, a blockbuster-potential drug will emerge from a process that started with a computer, then to bench, then to clinic. We will see within the next few years promising compounds enter the FDA pipeline.

Prediction 3: Pharma companies will step up M&A, investment and joint venture activity this year and in 2019 as they look to the world of software AI startups for new drug discovery capabilities. The tailwinds are there for tremendous value creation.

Photo: Natali_Mis, Getty Images

This post appears through the MedCity Influencers program. Anyone can publish their perspective on business and innovation in healthcare on MedCity News through MedCity Influencers. Click here to find out how.