A fresh viewpoint on drug discovery, pharma, and biotech

Become an author | Log in

Biopharma’s Hunt For Artificial Intelligence: Who Does What?

While Elon Musk and Mark Zuckerberg are arguing whether artificial intelligence (AI) is a long-awaited “blessing” or an existential threat to humanity, the progress in this exciting field is accelerating unabated and penetrating biopharma industry deeper.

The type of AI which scares some of the greatest minds, like Elon Musk and Stephen Hawking, is called “general artificial intelligence” -- the one which can “think” pretty much like humans do, and which can quickly evolve into a dangerous “superintelligence”. There is a notion that it will be invented in the nearest decades, but today we are definitely not there yet. The AI which is making headlines these days is a “narrow artificial intelligence”, a limited type of machine “intelligence” able to solve only a specific task or a group of tasks. It can’t go anywhere beyond specifics of the problem for which it is designed, so apparently, it will not hurt anyone in the nearest time. But now it can already provide fantastic practical results on those narrow tasks, like natural language processing, image recognition, controlling self-driving cars, and helping develop new drugs more efficiently. With the ability to find hidden and unintuitive patterns in vast amounts of data in ways that no human can do, AI represents a considerable promise to transform many industries, including pharma and biotech.  

The interest in AI-driven solutions for early stage drug discovery is growing steadily among biopharma leaders with a projected market volume reaching $10B by 2024 (for AI-based medical imaging, diagnostics, personal AI assistants, drug discovery, and genomics). The last couple of years were marked by a wave of new R&D collaborations between key biopharma players and AI-driven companies, primarily startups. Let’s see who is doing what in the biopharmaceutical AI landscape.

(Since most AI-driven companies use a mix of different approaches to solving drug discovery problems, the below classification is indicative).

AI for target based and phenotypic drug discovery

GlaxoSmithKline has just signed a $43 M drug discovery collaboration with U.K.-based AI-driven startup Exscientia to identify small molecules for ten selected targets across undisclosed therapeutic areas. Using a rapid “design-make-test” cycle Exscientia is able to design new molecules using AI-system, employing as well phenotypic and high content screening data, and assess their potency, selectivity and binding affinity towards specific targets. The projects will be heavily supported by Exscientia’s big-data resources—from its medicinal chemistry and large-scale bio-assays.

A month earlier Takeda announced a multi-year research partnership with AI-driven drug design company Numerate to develop new clinical candidates in oncology, gastroenterology, and central nervous system disorders. Numerate plans to apply AI-based modeling at every stage of the process -- from hit finding and expansion through lead design/optimization -- to absorption, distribution, metabolism, and excretion (ADME)/toxicity predictions. As stated on the website of Numerate, their AI-platform is able to work with data points obtained from different studies -- from high-content, low-throughput phenotypic assays as well as high-throughput screening, structure-based design, and traditional computational methods. Trained with versatile information, the AI-system can probe very large chemical spaces and identify the most promising drug candidates.

Another drug discovery focused company with advanced AI technology is Atomwise, which uses deep convolutional neural network AtomNet to empower structure based drug design. Atomwise was founded in 2015 and since then established a number of collaborations with high profile organization, such as pharma giant Merck, Stanford University, Duke University School of Medicine and so on. The company’s AI-based system can learn to recognize protein and ligand structures and interactions starting from the simplest features up to full objects, basically “teaching itself a chemistry course”. Further, it can model bioactivity of small molecules and chemical interactions and identify new molecules for the targets with previously unknown modulators. AtomNet outperformed traditional docking approaches, as was shown on benchmark examples.  

AI for biomedical data-driven drug discovery

Another opportunity for AI models to shine in the area of drug discovery is using biomedical data to draw unintuitive insights about drug candidates. Earlier this year, Santen Pharmaceuticals, a Japanese leader in the ophthalmic field, entered a strategic research collaboration with TwoXAR -- an AI-driven biopharmaceutical company. Instead of using molecular modeling techniques, TwoXAR works with real world biomedical data including gene expression measurements, protein interaction networks, and clinical records. By examining billions of points of information, TwoXAR’s AI platform is able to determine what is relevant, and what is noise, leading to a set of associations indicating the effectiveness of certain small molecules.

In the case with Santen, TwoXAR is aiming to find new drug candidates for glaucoma treatment and for that the company is screening large catalogs of molecules, associated with known data, such as protein structures, binding affinities etc. The data are then linked with molecular changes in glaucoma to derive unique disease-drug associations.

Another company focusing its AI-driven efforts on biology, rather than chemistry is Berg Health, advancing deep-learning screening of biomarkers from patient data and “multi-omic” modeling approaches. The types of data Berg is feeding to its AI algorithms includes not just genome data, but also the proteome, metabolome, and the lipidome of the biological samples to unravel the complex biological networks playing roles in diseases. That, in turn, can likely help identify medications for specific patient populations and, on the other hand, sift through the drug candidates that are likely to fail.

AI for polypharmacology discovery

Image credit: Kirschner Lab

While “one target one disease” has been a dominating paradigm in drug discovery for years, it is becoming obvious that many diseases are too complex to be efficiently cured within this paradigm. A multitarget drug discovery approach is a promising way to make more efficient medicines.

With this in mind, Sanofi put a $274 M deal with AI-driven Exscientia in 2017 to discover and develop bispecific small molecules that treat diabetes and its comorbidities. Exscientia’s role will be to come up with pairs of targets, related to glucose control, NASH, weight management and other diabetes-related areas, and generate bi-specific small molecule ligands using AI-based platform.

Similar multitarget strategy was pursued in the other Exscientia’s research collaboration with Evotec in 2016 to discover and develop first-in-class bispecific small molecule immuno-oncology therapies. As in the case with Sanofi, Exscientia will provide value to Evotec via its AI-driven platform to purposely design bispecific small molecules that can address multiple targets through a single drug.

AI for drug repurposing programs

Drug repurposing is one of the gold mines for AI-based technologies to drive value since a lot of data is already known about the drug in question. Repurposing previously known drugs or late-stage drug candidates towards new therapeutic areas is also a desired strategy for many biopharmaceutical companies as it presents less risk of unexpected toxicity or side effects in human trials, and, likely, less R&D spend.    

An illustrative example is a recent R&D partnership between Sanofi and an emerging AI-driven biotechnology company Recursion Pharmaceuticals in 2016 with the purpose to identify new uses for Sanofi’s clinical stage molecules across dozens of genetic diseases. The Recursion’s approach is a “target-agnostic” one and is based on cellular phenotyping via image analysis using computer vision. Thousands of morphological measures are thus extracted at the level of individual cells and large catalogs of molecules are screened for the ability to “fix” phenotypic defects associated with each disease. Under the agreement, Sanofi will provide Recursion with a number of small molecules, and Recursion will screen them across its rapidly expanding library of genetic disease models and use machine learning technology to reveal promising new indications.

Astellas Pharma Inc. signed a research deal with big data-driven bioinformatics company NuMedii to conduct drug repurposing projects using machine learning techniques. NuMedii’s big data resource includes hundreds of millions of human, biological, pharmacological and clinical data points, normalized and annotated. The company then uses neural network-based algorithms to find novel drug candidates, and biomarkers predictive of diseases, and repurpose existing drugs or drug candidates towards other medical indications.

Another example of using AI-algorithms for drug repurposing programs was demonstrated by Insilico Medicine, which developed deep learning tools for predicting pharmacological properties of drugs and drug repurposing using transcriptomics data. Using its AI-expertise, the company also has plans to advance drug discovery research in the area of aging-related diseases. Insilico Medicine has partnership relations with Novartis, Johnson & Johnson and other major industry and research organizations.

AI for analyzing research literature, publications, and patents

Image credit: Fujitsu

Reading, clustering and interpreting large volumes of textual data is among the most successful use cases for AI-based algorithms. It comes in handy for life sciences industry since the number of research publications in the field is growing enormously and it is hard for researchers to sift through vast amounts of data arriving on daily basis to validate or discard research hypotheses.

A research collaboration between pharmaceutical giant Pfizer and IBM’s Watson for Drug Discovery to tackle immuno-oncology was announced last year and became one of the most covered news stories of AI application in biopharma sector. This collaboration was aimed to bring the power of AI-driven super-computer for accelerating analysis and tests of hypotheses by researchers at Pfizer using “massive volumes of disparate data sources” that include more than 30 million sources of laboratory data reports as well as medical literature.

A notable company which uses AI for working through published research literature and patents is UK-based BenevolentAI. According to the company’s representative James Chandler: “A new scientific paper is published every 30 seconds and there are 10,000 updates to PubMed every day”. Navigating through all this information to draw meaningful insights about drug candidates is where AI-based algorithms become indispensable -- this is the type of things BenevilentAI does.

In Nov 2016 BenevolentAI signed a licensing deal with Janssen Pharmaceuticals, a Johnson & Johnson company, for a sole right over a series of small molecule candidates and patents. The company plans to use its AI-driven platform to accelerate drug discovery and likely find new therapeutic indications for the selected small molecules. According to the license agreement, BenevolentAI can develop, manufacture and commercialize those novel drug candidates in all indications.

Conclusions: R&D outsourcing and M&A will thrive in biopharma

With an increasing interest in AI-driven technologies among the leading biopharmaceutical companies, a strategic focus of pharma and biotech businesses will be further shifting towards R&D outsourcing and M&A activity as means to quickly get access to the required expertise and know-hows. Complex nature of AI-based technologies, a need for costly and sophisticated IT infrastructure, a fast pace of progress in the field, and a relative scarcity of highly skilled data science specialists to support specialized machine learning research -- these are some of the key drivers of the ascending outsourcing trend.

You may also be interested to read:


  • Ed Addison 2017/08/08, 11:32 AM

    A couple of links provide additional information not here addressed:

    • BiopharmaTrend 2017/08/14, 11:44 AM

      Thank you for the valuable updates! These links will be included in the coming reviews.

  • Aman 2017/08/10, 20:35 PM

    Thanks, interesting read.

    I am surprised that InveniAI (off shoot of a company called BioXcel Corporation) never made it your list. InveniAI has been successful not only with Pharma partnerships (Takeda, Alnylam, Axcella, Centrexion all announced in public domain) but also in spinning out drug companies, "BioXcel Therapeutics, 2 Phase II programs", spin out with another large pharma.


    • BiopharmaTrend 2017/08/14, 11:48 AM

      Dear Aman, thank you for this additional and valuable information! I will include InveniAI in the upcoming review update at the end of the year. This commentary will make the next review much more informative, indeed. Regards, Andrii

  • Joe Donahue 2017/08/15, 23:30 PM

    Great article. Thank you. This is a rapidly evolving space - both with horizontal AI technologies being applied to pharma discovery as well as a wave of companies - Vyasa is one that didn't make your list - that are focused on on leveraging their deep knowledge of the life sciences vertical.

    • BiopharmaTrend 2017/08/16, 10:39 AM

      Dear Joe, thank you for updating the current list with a valuable contribution. In fact, commentaries here are valuable for our readers and sometimes even more insightful than the article itself.


Leave a Reply

Your email address will not be published. Required fields are marked *