top of page
Business Meeting

Tech-First Companies Take the Lead in AI Drug Discovery

In recent news, Recursion (NASDAQ: RXRX), a prominent clinical stage TechBio firm, has announced the signing of agreements to acquire Cyclica and Valence, two companies with expertise in AI-enabled drug discovery. These acquisitions strengthen Recursion's position in computational chemistry, machine learning, and artificial intelligence, enhancing its technology-enabled drug discovery capabilities in the biopharma industry.

Cyclica, based in Toronto, has developed two innovative products in the computational chemistry domain, MatchMaker™ and POEM™ (Pareto Optimal Embedding Model), both of which will be integrated into the RecursionOS. MatchMaker™ is a deep learning engine that leverages AI to predict the polypharmacology of small molecules for drug discovery. POEM™ is a similarity-based property prediction model that provides a more accurate and comprehensive measure of molecular similarity, setting it apart from other AI prediction models.

Naheed Kurji, CEO and Co-Founder of Cyclica, stated that integrating Cyclica's proteome-wide prediction capabilities into Recursion's data ecosystem will result in one of the most extensive and purpose-built biological and chemical datasets in the drug discovery space.

Valence, a Montréal-based company located at Mila, the world's largest deep learning research institute, focuses on harnessing the power of deep learning for drug discovery. The firm has been a pioneer in applying low-data learning to drug design, enabling the development of differentiated small molecules with improved properties and functionality from datasets unsuitable for traditional deep learning methods.

Daniel Cohen, CEO and Co-founder at Valence Discovery, expressed enthusiasm about integrating Valence's AI-based chemistry engine into Recursion's diverse and data-rich operating system, which he believes will help unlock the true potential of AI-first digital chemistry and drug discovery.

Upon acquisition, Valence will join forces with Recursion's Montréal deep learning research office, transforming into an artificial intelligence and machine learning research center led by Daniel Cohen, with continued advisory from Yoshua Bengio.

Recursion will acquire Cyclica for $40 million and Valence for $47.5 million, with both acquisitions expected to be completed in the second quarter of 2023, subject to closing conditions. The purchase price will be payable in the form of shares of Recursion Class A common stock, shares of a subsidiary of Recursion exchangeable for shares of Recursion’s Class A common stock, and the assumption of certain outstanding Valence and Cyclica options.

Building drug discovery platforms

Recursion is a vivid example of a new generation of “digital biotechs” -- companies which are built around AI-driven highly integrated and data-centric R&D workflows, often presented in a form of research platforms. Such companies are strikingly different, business model-wise, from traditional drug discovery and biotech companies, often centered around a particular therapeutic asset in development.

Conceptually, the term "platform" signifies a holistic and interconnected system that amalgamates an array of tools, technologies, and algorithms, ultimately expediting and refining the drug development pipeline. Various components of a platform work in concert to process copious amounts of biological, chemical, and clinical data, driving innovation in the pharmaceutical sector.

For instance, the Recursion Operating System (OS) offers a novel perspective on drug discovery by attempting to treat it as a search problem that can be methodically addressed. Central to the company's objectives, this integrated and multifaceted system is constructed to generate, analyze, and draw insights from large-scale biological and chemical datasets, with the aim of expediting the drug discovery process.

Three essential components form the backbone of the Recursion OS. Firstly, the Infrastructure Layer comprises both hardware and software, ensuring a stable foundation for the efficient operation of the system. Secondly, the Recursion Data Universe serves as a vast repository for diverse datasets, offering a wealth of information for researchers and data scientists to work with. Lastly, the Recursion Map is a collection of proprietary tools dedicated to discovery, design, and development in the drug discovery process.

Other AI in drug discovery companies building R&D platforms

There are a plethora of other AI drug discovery companies which operate sophisticated R&D platforms.


For example, Exscientia, a leading UK-based AI drug discovery company, has developed the CentaurAI platform, a patient-first AI system aimed at creating medicines with an increased likelihood of success.

One of the workflows in Exscientia’s AI drug discovery platform

By integrating high-precision data from patient tissue with the CentaurAI platform, the company is able to streamline target selection, precision design and experimentation, as well as enhance clinical assessments. The AI-driven platform identifies emerging opportunities in target and disease space by combining global genetic data, literature, and primary patient tissue readouts. Exscientia's innovative approach includes phenotypic discovery, druggability, and precision design, among other elements, to optimize drug development processes while utilizing data-agnostic and generative design techniques. This advanced AI drug discovery platform is revolutionizing the pharmaceutical industry and setting new standards for precision medicine, as evidenced by the landmark EXALT-1 clinical study, which demonstrated improved patient outcomes using functional precision-oncology.

Over several years, Exscientia demonstrated an impressive performance of its AI drug discovery platform. In particular, its partnership with Celgene, which was later acquired by Bristol-Myers Squibb (BMS), led to the development of EXS4318, a selective Protein kinase C (PKC) theta inhibitor. The collaboration expanded to include immunology and oncology candidates, with BMS increasing its potential payments to Exscientia to over $1.3 billion.

Another Exscientia AI-inspired drug candidate, EXS21546, is being co-developed with Evotec as an anti-cancer immunotherapy and is currently in Phase I/II trials. The company is also working on several early discovery oncology candidates in collaboration with Sanofi, GT Apeiron, EQRx, and Huandong. In addition, Exscientia's pipeline includes inflammation and immunology candidates, a hypophosphatasia candidate co-owned with Rallybio, a coronavirus candidate developed with the Bill and Melinda Gates Foundation, and a late discovery psychiatry candidate in partnership with Blue Oak Pharmaceuticals.

Exscientia has also used its AI drug discovery platform to co-develop a Phase I clinical candidate, DSP-0038, for Alzheimer's disease psychosis with Sumitomo Pharma.


One more UK-based company BenevolentAI developed The Benevolent Platform™, an AI drug discovery platform that offers a comprehensive view of disease biology by integrating over 85 data sources from various domains and data types. This platform helps scientists break down silos across therapeutic areas, connecting shared mechanisms across diseases.

Components of The Benevolent Platform™

The Benevolent Platform™ processes massive quantities of biomedical data using custom-engineered pipelines and machine learning models, which extract biomedical entities and infer relationships. This information is stored in the Knowledge Graph, providing a proprietary integrated view of biomedical data that supports discovery and decision-making.

In early 2020, BenevolentAI scientists used the Knowledge Graph to discover a COVID-19 treatment, which is now approved by the US FDA. The platform's AI models mine the Knowledge Graph to identify novel insights and relationships, uncovering previously unconsidered drug targets.

The Benevolent Platform™ can also be applied to antibodies and other biologic agents, expanding the pool of possible drug targets. By integrating disease traits, genetics, and genomics into the Knowledge Graph, the platform generates endotype-specific target predictions that maximize the chances of clinical success. The AI tools provided by the platform enable scientists to make data-driven decisions and select only the most promising targets for wet lab experiments. Through the combined capabilities of the AI drug discovery platform, scientific expertise, and wet lab facilities, BenevolentAI has rapidly built a substantial portfolio of best-in-class and first-in-class drug candidates.

Insilico Medicine

Hong Kong-based Insilico Medicine strives to accelerate drug discovery and development in three key areas: disease target identification, generation of novel molecules data, and predicting clinical trial outcomes. Their PHARMA.AI suite includes PandaOmics, Chemistry42, and inClinico. PandaOmics allows users to access processed OMICs data for easier interpretation and uses a proprietary pathway analysis approach, iPanda, to provide comprehensive analysis. It also prioritizes and filters target hypotheses based on multiple scores derived from text and omics data. Chemistry42 assists in defining rules for molecule properties and optimizes compounds with physico-chemical parameters, binding scores, and drug-likeness features. Lastly, inClinico is a data-driven platform for predicting the probability of success for individual clinical trials, utilizing vast amounts of data related to targets, diseases, trials, and scientists involved in the studies.

Insilico Medicine demonstrated a track record of successful validations of their AI drug discovery platform, including accelerated results for various indications, such as fibrosis, inflammation, and oncology, at a fraction of the typical R&D cost. Since 2021, the team has delivered 9 preclinical candidates discovered and designed using its AI drug discovery platform, including one for the QPCTL immuno-oncology target in partnership with Fosun Pharma.

In early 2023, Insilico Medicine announced positive topline results for Phase 1 clinical trials of its AI-designed novel drug candidate INS018_055 for a novel target in idiopathic pulmonary fibrosis (IPF). The positive Phase I data paves the way for further evaluation of the drug's efficacy in IPF patients in a Phase II trial.

Notably, Insilico Medicine recently launched Life Star, a 6th generation Intelligent Robotics Drug Discovery Laboratory, in Suzhou BioBAY Industrial Park. The Intelligent Robotics Lab forms a closed loop by combining the company’s AI drug discovery platform Pharma.AI with fully automated biological experiment modules.


BPGbio, a clinical-stage AI drug discovery company, has recently acquired assets from BERG, LLC, including their proprietary Interrogative Biology® Platform. This platform, developed by BERG since 2006, utilizes multi-omics data from over 100,000 samples, advanced AI software, and the powerful Frontier supercomputer at Oak Ridge National Laboratory.

With a focus on hypothesis-free, data-driven research, the platform delves deep into molecular phenotyping beyond traditional genomics, incorporating proteomics, lipidomics, and metabolomics. This approach has applications in various therapeutic areas, such as oncology, neurology, and metabolic diseases, and has been used for target discovery, mechanism of action discovery, drug repositioning, and biomarker discovery. BPGbio's diverse clinical pipeline, now bolstered by BERG's technology, is led by BPM 31510, a novel drug candidate targeting the metabolism of cancer cells and showing potential in the treatment of pancreatic cancer. As an AI drug discovery company, BPGbio's acquisition of the Interrogative Biology® Platform signals a promising step forward in utilizing AI-driven solutions for drug discovery and development.

In 2018, the FDA awarded orphan drug designation to BPM 31510 for its potential in treating pancreatic cancer and epidermolysis bullosa (EB).

Deep Genomics

In 2019, Canadian-based Deep Genomics made headlines with discovering a novel target and a novel RNA therapeutics candidate for rare Wilson disease using their AI Workbench platform -- all within 18 months of initiating target discovery effort. In 2021, the company raised $180 million from a number of venture capital partners. There has not been much news from the company since then, though.

"AI-first" drug discovery companies reimagine pharmaceutical research

What do all these AI drug discovery companies and their recent successes have in common? A closer look at their development path, a track record of publications or announcements over the last several years reveal a quite notable distinction from a “traditional” drug discovery company -- they were all built as “AI-first” projects from the very beginning or adopted AI technologies in early stages of building organizational R&D workflows. Basically, they were growing as tech companies in the biomedical space, unlike many “wet-lab-born” drug discovery companies out there.

For example, Insilico Medicine started as a deep learning drug discovery company and was initially focused on generative models for drug design. Over the years, the company managed to gradually build an “end-to-end” drug discovery system, including everything from concept generation and target discovery to drug design and even clinical trial modeling. While the platform includes hundreds of components and models, working in an ensemble, a notable part of it is so-called generative adversarial networks (GANs) -- an innovative class of machine learning frameworks designed by Ian Goodfellow and his colleagues in June 2014, and pioneered by Insilico Medicine for applications in drug design.

BenevolentAI early in the development built its integrated Benevolent platform with Knowledge Graph being at the heart of target discovery and drug repurposing. Exscientia developed its end-to-end AI platform Centaur Chemist which works in a combination with human expertise. Recursion Pharmaceuticals was built as a digital biology company in the first place, with all drug design successes revolving around its AI-driven biology analytics and cell image recognition capabilities. Deep Genomics developed AI Workbench as a cornerstone of its research process.

Moderna’s success in AI drug discovery and vaccine development

Probably, the most well-known to the general public example of a company, which managed to realize the “AI-first strategy” at scale, comes from the area of mRNA-based vaccine development and therapeutics.

Moderna Therapeutics has demonstrated tremendous growth over just several years, and it had artificial intelligence and “tech-first” strategy in its business model from the very beginning. The company managed to come up with an mRNA vaccine within just months -- a hard task. The success came from a combination of basic biology breakthroughs, delivery technology, and carefully crafted digital systems and automation workflows to do it all efficiently at speed and scale.

As Marcello Damiani, Chief Digital Officer at Moderna wrote in 2017 white paper, “We rely on digitization to ensure seamless integration across the ecosystem, enable the ability to share and access data, permit the capability to scale, satisfy ever-increasing demands for research mRNA for preclinical and GLP toxicology studies, as well as GMP mRNA to supply an expanding number of clinical studies.”

“Our platform comprises five key elements – Chemistry, Bioinformatics, mRNA Engineering, Process, Formulation”

“Digitization is the backbone upon which our platform is built. It is both an enabler of our science and core to our science”

“This enterprise-wide focus on digitization has positioned Moderna to execute against our strategy, while also yielding a distinct competitive advantage. Six years since commencing operations, Moderna has amassed a pipeline with a breadth, speed, and scale not common in our industry”.

Drawing parallels among companies like Moderna, Recursion Pharmaceuticals, Exscientia, Insilico Medicine, and alike, it becomes apparent that a new type of organizations emerged in the pharma industry -- “tech-first” companies, with more flexible business models, faster R&D workflows, more optimal data management. This is the era of data-driven research and industrialized drug discovery.

AI in drug discovery: evolution, not a revolution

A recent report by BiopharmaTrend summarizes some of the leading clinical stage drug candidates, developed using artificial intelligence -- out of dozens of other such candidates currently in clinical development. The application of “AI-first” drug discovery strategy and technology integration allows a number of companies, such as Exscientia, BenevolentAI, Insilico Medicine and others, to demonstrate substantial acceleration of preclinical research timelines, down to 18 months and less, for the entire route from idea to preclinical candidate nominations.

However, the advent of artificial intelligence in the biomedical field is not as immediately disruptive as it is in other industries, such as text translation, video generation and ChatGPT.

Biology is much harder and it takes a lot of time to confirm or disprove AI model prediction outcomes, which are not always as expected. Already today it is clear that AI is extremely promising, but will not be able to disrupt pharmaceutical research in any instance. It will most likely be a gradual evolution of research practices and technological stacks, as well as business models of drug discovery companies.

For example, among various accomplishments, such as the swift discovery of baricitinib through AI-assisted drug repurposing and other recent achievements, BenevolentAI recently reported mixed topline results from its Phase IIa trial examining AI-derived drug candidate BEN-2293 in patients with mild-to-moderate atopic dermatitis (AD). BEN-2293 is a selective inhibitor of the three tropomyosin-related kinases (Trk) receptors (TrkA, TrkB, and TrkC), often referred to as a pan-Trk inhibitor, and is designed for topical application.

The study's primary endpoint assessed the safety and tolerability of BEN-2293, while secondary outcomes evaluated the percentage of patients achieving an improvement in the Eczema Area and Severity Index (EASI) and pruritus Numerical Rating Scale (NRS).

Although BEN-2293 was safe, well-tolerated, and met the trial's primary endpoint, it did not demonstrate a statistically significant effect on EASI or NRS endpoints among the participants in the treatment arm.

We will be able to assess the current impact of AI in drug discovery when more of the AI-inspired drug candidates currently in early clinical development advance to further stages (or fail). In the meantime, we at BiopharmaTrend created the BPT20: Artificial Intelligence in Drug Discovery Productivity Index, with the idea of tracking the progress in this exciting and rapidly evolving field.

August 7, 2023

Andrii Buvailo, Ph.D.

Co-Founder, Director, BiopharmaTrend

Andrii Buvailo is a pharmaceutical industry analyst and writer, focusing on

emerging companies (startups), technologies and trends in drug discovery, and

R&D outsourcing. He received a master’s degree in Inorganic Chemistry and

a PhD in Physical Chemistry from Kyiv National Taras Shevchenko University.

His articles were published on, and market research reports

were referenced by some of the leading life science organizations. He also

participated in numerous scientific projects in Ukraine, Belgium, Germany, and

the United States (DAAD, Horizon 2020, NATO, CRDF grants), and published in

high-impact research journals.


bottom of page