It’s Been a Decade of AI in the Drug Discovery Race. What’s Next?

| Apr. 3, 2024

The year 2012 is often regarded as the beginning of the deep learning revolution.

That year, Alexnet, the deep learning model proposed in the research paper Imagenet Classification with Deep Convolutional Neural Network by Alex Krizhevsky and his colleagues, won the large-scale visual recognition challenge ImageNet -- completely dominating over other machine learning competitors.

Another famous event of that year was spontaneous identification of a cat by another deep neural network -- in millions of random YouTube videos.

Scientists at Google's mysterious X lab built a neural network of 16,000 computer processors with one billion connections, and it was never told to look for cats, nor was it told what a cat was. Artificial intelligence learned that by itself and flagged the pattern of pixels corresponding to a cat’s face.

While the concept of deep learning was known since the 1960s, the technology only started manifesting itself as a practically viable thing in the 2010s, mainly due to sufficient progress in computational power allowing it to train such complex models.

Additionally, 2000-2010s were marked by the increasing availability of big datasets for training purposes, such as ImageNet, and the rise of computational infrastructure, like Elastic Compute Cloud by Amazon -- other important drivers of the deep learning revolution which started unfolding in 2010-2012.

The drug pipeline progress of some of the leading AI in drug discovery companies over the years. See below for details.

Historical trajectory of AI in drug discovery

It is no wonder then, that more than 80% of more than 500 AI-augmented drug discovery and development startups and scaleups recorded in BiopharmaTrend database “The State of AI in the Biopharma Industry” were founded around 2012 and later, around the time deep learning became a major trend.

2012

Some of the early AI companies, founded in 2012, included Atomwise (structure based small molecule screening), Exscientia (target discovery and small molecule design), AbCellera (genomics-driven antibody design), and Flatiron Health (a clinical data integrator and analytical powerhouse for clinical research in oncology).

2013

A year later, other notable AI-driven players emerged, including BenevolentAI, Cyclica, and Recursion Pharmaceuticals -- in the small molecule drug discovery space, and Zymergen in biotech.

2014

But the progress in the artificial intelligence space accelerated, and the year 2014 saw another major milestone, later called by Yann LeCun, Facebook's chief AI scientist at that time, “the coolest idea in deep learning in the last 20 years.”

In June 2014, Ian Goodfellow published a seminal paper introducing Generative Adversarial Networks (GANs). Not only did this breakthrough concept transform the area of generative AI, but it also won his inventor a place in history as The GANfather: The man who’s given machines the gift of imagination.

This breakthrough stimulated the next spin of innovation and drug discovery entrepreneurship, including probably the most notable AI-driven company founded in the year 2014 -- Insilico Medicine.

Co-founded by Dr. Alex Zhavoronkov, Insilico Medicine pioneered the application of deep learning for small molecule drug design and later -- target discovery. The company also incorporated GAN technology, and later built a number of GAN-based computational tools for drug discovery, including DruGAN for fingerprints, ORGAN for SMILEs, etc.

As per information I have received by email from Insilico Medicine, most of their targets are novel or have sufficient novelty aspect, and they managed to discover both novel targets and molecules for those targets relying mostly on their Pharma.AI engine (see Table 1 below for detailed pipeline information).

2015

This year, a prominent AI company XtalPi was founded on the MIT campus and in China, focusing from its inception on combining quantum physics, AI, cloud computing, and large-scale clusters of robotic workstations.

Interestingly, the company initially disclosed an internal pipeline, including at least 10 early stage discovery programs for a number of indications. However, at present, the company is focusing exclusively on partnerships and contract research services.

As the a-1 filing documents indicate:

“We have entered into a number of collaborations with biotechnology and pharmaceutical companies and academic institutions under which our collaborators pursue research in a number of therapeutic areas, such as oncology, neurology, respirology, and inflammatory diseases. In some cases, we retain at least partial ownership in the pipeline programs, typically in the double-digit percentage range, of the programs pursued under these collaborations. We are not responsible for advancing their pre-clinical development beyond generation of pre-clinical candidates.”

As per company email for BiopharmaTrend, XtalPi never intended to pursue these projects in the same sense the other companies build up their pipelines. When they first launched their one-stop drug discovery business, the company started several proof-of-concept discovery projects in-house to showcase capabilities to clients and accumulate R&D data.

Currently, XtalPi offers drug discovery solutions (both in small molecules and antibodies) that are powered by AI and (robotics) wet lab, which play a more prominent role in the zero-to-one innovation process of identifying novel molecules for clinical candidates.

2015-2019

Many notable AI-focused companies were founded in the coming years to tackle various aspects of drug discovery and development, including Insitro, Relay Therapeutics, Valo Health, Verge Genomics and others.

We should mention one notable exception -- New York-based AI drug discovery company Schrödinger, which was founded in 1990, well before the deep learning era.

In the early years, the company was known as a well-established developer of cheminformatics solutions and software for drug research. Over the last decade, Schrödinger picked the AI trend as well and substantially augmented its product offering, launched novel AI-based tools, and created an internal drug candidate pipeline eventually, with two Phase 1 assets (Table 1).

Trying to comprehend the market of AI in drug discovery and development, it should be noted that a broad category of the 500+ AI startups recorded in BiopharmaTrend report are not developing own drug candidates, they have no internal pipelines.

For instance, many successful companies on the list, like CytoReason (AI-enabled disease modeling) or BenchSci (AI-driven search engine for scientific reagents), contribute meaningfully to the advent of artificial intelligence in the pharma and biotech industries. But they focus on services, software licensing, and R&D partnerships.

Two other broad categories of AI startups in the pharmaceutical industry include those in the drug repurposing space, like Healx and Lantern Pharma, and AI vendors in the clinical trial space, like Medidata.

In this report, we are exclusively focused on the companies that can design and advance drug candidates de novo (mostly, small molecules), and have internal pipelines.

A Wave of AI-designed Drug Candidates Hits The Breakwater

While hundreds of AI startups were founded since 2012, and raced towards building various systems for drug discovery, the wave of first preclinical and clinical candidates out of those efforts came later.

For instance, Insilico made first headlines in 2019, having reached a notable proof-of-concept milestone where they predicted a molecule for a well-known target called DDR1 in just 21 days — and successfully validated prediction in vitro and in vivo.

In the same year, Toronto-based Deep Genomics announced “industry’s first AI-discovered therapeutic candidate” DG12P1, for Wilson disease. Deep Genomics’s AI platform pinpointed a specific genetic mutation and helped design a compound to correct it, all within 18 months.

2020-2022

Around 2020-2022, a series of preclinical candidate nominations included AI-discovered targets and molecules by BenevolentAI, Exscientia, Insilico Medicine, Deep Genomics, and others, and around 2022 several dozens of AI-generated molecules were in clinical trials.

In order to comprehend the advance of multiple AI-discovered molecules over the years, we have conducted a historical pipeline analysis of some of the most developed AI in drug discovery platforms, including (alphabetically):

BenevolentAI, Exscientia, Insitro, Insilico Medicine, Recursion Pharmaceuticals, Relay Therapeutics, Schrödinger, Verge Genomics, and Valo Health.

We did not include partner programs in this analysis.

Disclaimer

———	Program	Ownership	Target	Indication	2019	2020	2021	2022	2023
Table 1 DISCLAIMER: Historical data sourced from archived public sources, including Webarchive service (see references 1-25), the data is NOT provided by the companies, and may have evolved.
BenevolentAI	BEN-8744	Whole	PDE10	Ulcerative Colitis			Discovery	Preclinical	Phase 1
	BEN-28010	Whole	CHK1	Glioblastoma Multiforme			Discovery	Preclinical	Preclinical
	BEN-34712	Whole	RAR⍺β	ALS				Discovery	Preclinical
	-	Whole	-	Parkinson's disease				Discovery	Discovery
	-	Whole	-	Fibrosis				Discovery	Discovery
	Partnered Program	Co-owner w/ AstraZeneca	-	Chronic Kidney Disease					Discovery
	-	Co-owner w/ AstraZeneca	-	Idiopathic Pulmonary Fibrosis					Discovery
	-	Co-owner w/ Merck	-	Oncology					Discovery
	-	Co-owner w/ Merck	-	Neurology					Discovery
	-	Co-owner w/ Merck	-	Immunology					Discovery
	BEN-2293	Whole	TrkA, TrkB, and TrkC	Atopic Dermatitis	Discovery	Preclinical	Phase 1	Phase 2	Phase 2`
	BEN-9160	Whole	Bcr-Abl	ALS			Discovery	Preclinical	unknown
	-	Whole	-	Inflammatory bowl disease (IBD)				Discovery	unknown
	-	Whole	-	Antiviral				Discovery	unknown
	-	Whole	-	Oncology				Discovery	unknown
	-	Whole	-	Oncology				Discovery	unknown
	-	Whole	-	NASH				Discovery	unknown
	-	Whole	-	Oncology				Discovery	unknown
	-	Whole	-	Parkinson's disease				Discovery	unknown
	-	Whole	-	Inflammation				Discovery	unknown
Exscientia	EXS21546	Majority, w/ Evotec	A2aR	High Adenosine Signature Cancers		Preclinical	Phase 1	Phase 1/2	Phase 1/2`
	-	Whole	HPK1	Immuno-Oncology			Discovery	unknown
	EXS74539	Whole	LSD1	Oncology, AML, SCLC			Discovery	Discovery	Preclinical
	EXS73565	Whole	MALT1	Oncology, Hematology			Discovery	Discovery	Preclinical
	-	Whole	-	Oncology			Discovery	unknown
	-	Whole	-	Oncology			Discovery	unknown
	-	Whole	-	Oncology			Discovery	unknown
	-	Whole	-	Oncology			Discovery	unknown
	-	Whole	Mpro	COVID-19			Discovery	Preclinical	unknown
	-	Whole	-	Anti-infective			Discovery	unknown
	-	Whole	NLRP3	Inflammation and Immunity			Discovery	Preclinical	unknown
	GTAEXS617	Co-owner w/ Apeiron	CDK7	Transcriptionally addicted cancers			Preclinical	Preclinical	Phase 1/2
	EXS4318	Out-licensed, BMS	PKC-theta	inflammatory and immunologic diseases		Preclinical	Preclinical	Preclinical	Phase 1
	-	Co-owner	-	Oncology			Discovery	Discovery	unknown
	-	Co-owner	ENPP1	Oncology			Discovery	Preclinical	unknown
	-	Co-owner	ENPP1	HPP			Discovery	Preclinical	unknown
	-	Co-owner	-	Inframmation and immunity			Discovery	Discovery	unknown
	-	Co-owner	-	Inflammation and Immunity			Discovery	Discovery	unknown
	-	Co-owner	-	Oncology			Discovery	Discovery	unknown
	-	Co-owner	-	Oncology			Discovery	Discovery	unknown
	-	Co-owner	-	Psychiatry			Discovery	Preclinical	unknown
Insilico Medicine	INS018_055	Whole	TNIK	IPF		Discovery	Preclinical	Phase 1	Phase 2
	-	Whole	TNIK	Kidney fibrosis			Discovery	Preclinical	Preclinical
	-	-	TNIK	Skin Fibrosis			Discovery	Discovery	unknown
	-	Whole	TNIK	IPF (inhalable)			Discovery	Discovery	Preclinical
	ISM012	Whole	PHD1/2	Anemia of Chronic Kedney Disease			Discovery	Preclinical	Phase 1
	-	Whole	PHD1/2	Infammatory bowl disease (IBD)			Discovery	Preclinical	Phase 1
	ISM8207	Co-owner w/ Fosun	QPCTL	Immuno-oncology			Discovery	Preclinical	Phase 1
	-	Co-owner w/ Fosun	-	Diabetic neuphropathy, FSGS			Discovery	unknown
	ISM3312	-	3CLpro	COVID-19			Discovery	Preclinical	Phase 1
	ISM3412	Whole	MAT2A	MTAP-/-cancer			Discovery	Preclinical	Preclinical
	ISM9274	Whole	CDK12/13	Solid tumors			Discovery	Discovery	Preclinical
	ISM027	Whole	cMYC	Solid tumors					Discovery
	ISM022	Whole	CDK8	AML, Solid tumors			Discovery	Discovery	unknown
	ISM023	Whole	PARP7	Solid tumors			Discovery	Discovery	unknown
	ISM3091	Out-licensed, Exelixis	USP1	BRCA-mutant cancer			Discovery	Preclinical	Phase 1
	ISM5939	Whole	ENPP1	solid tumors			Discovery	Discovery	Preclinical
	ISM5043	Out-licensed, Menarini	KAT6	ER+/HER2-breast cancer					Preclinical
	ISM4525	Whole	DGKA	Solid tumors					Preclinical
	ISM8001	Whole	FGFR2/3	Solid tumors					Preclinical
	ISM016	Whole	NLRP3	Gout flare			Discovery	Discovery	Discovery
	ISM6331	Whole	TEAD	Solid tumors					Preclinical
Recursion Pharmaceuticals	REC-4881	Whole	MEK1 and MEK2	Familial Adenomatous Polyposis		Preclinical	Phase 1	Phase 1	Phase 2
	REC-3599	Whole	PKC and GSK3ß	GM2 Gangliosidosis		Preclinical	Phase 1	terminated
	REC-2282	Whole	HDAC	Neurofibromatosis Type 2		Preclinical	Phase 1	Phase 1	Phase 3
	REC-994	Whole	antioxidant, no specific target	Cerebral Cavemous Malformation		Preclinical	Phase 1	Phase 1	Phase 2
	REC-3964	Whole	-	Clostridium Difficile Colitis		Discovery	Preclinical	Preclinical	Phase 1
	-	Whole	-	Neuroinflammation			Discovery	Discovery	unknown
	-	Whole	-	Batten Disease			Discovery	unknown
	-	Whole	-	Charcot-Marie-Tooth Disease Type 2			Discovery	Discovery	unknown
	-	Whole	-	Immune Checkpoint resistance in STK11-NSCLC			Preclinical	Preclinical	unknown
	-	Whole	-	Oncoloty			Discovery	Discovery	Discovery
	25 programs	-	-	Various			Preclinical	unknown
	REC-4881	Whole	-	AXIN1 or APC Mutant Cancers					Phase 1
	-	-	-	Pulmonary Arterial Hypertension				Preclinical	unknown
	-	Whole	-	-				Preclinical	unknown
	Immunotherapy Target Alpha	Whole	-	Oncology				Discovery	Discovery
	Immunotherapy Target Beta	Whole	-	Oncology				Discovery	unknown
	-	Whole	-	Hepatocellular Carcinoma				Discovery	unknown
	-	Whole	RBM39	HR-proficient Ovarian Cancer RBM39					Preclinical
	Immunotherapy Target Delta	Whole	-	-					Preclinical
Relay Therapeutics	RLY-4008	Whole	FGFR2 (mutant+WT)	FGFR2-altered cholangiocarcinoma (CCA)	Discovery	Phase 1	Phase 1	Phase 1	Phase 1/2
	RLV-PI3K1047 (RLY-5836)	Whole	PI3Kα	-			Discovery	Preclinical	Phase 1
	RLY2608	Whole	PI3Kα	solid tumors with a PI3Kα mutation				Phase 1	Phase 1
	-	Whole	PI3Kα	-				Discovery	unknown
	RLY-2139	Whole	CDK2	Oncology			Discovery	Discovery	unknown
	-	Whole	ERα	-					unknown
	GDC-1971	Co-owner w/ Genentech	SHP2	Cancers, expand into multiple combination	Preclinical	Phase 1	Phase 1	Phase 1	Phase 1
	-	Whole	-	Oncology			Discovery	Discovery	unknown
	-	Whole	-	Oncology			Discovery	Discovery	unknown
	-	Whole	-	Genetic disease			Discovery	Discovery	unknown
	-	Whole	-	Genetic disease			Discovery	Discovery	unknown
Schrödinger	SDGR3 (SGR-1505)	Whole	MALT1	Relapsed or refractory B-cell lymphoma, chronic lymphocytic leukemia	Discovery	Discovery	Preclinical	Phase 1	Phase 1
	SDGR1	Whole	CDC7	Esophagial and Lung Cancers,	Discovery	Discovery	Discovery	unknown
	SGR-2921	Whole	CDC7	Hematological cancers and solid tumors				Preclinical	Phase 1
	SDGR2	Whole	WEE1	Ovarian, Pancreatic, Breast and Lung Cancers	Discovery	Discovery	Discovery	unknown
	SGR-3515	Whole	WEE1/MYT1	Solid tumors				Discovery	Preclinical
	SDGR4	Co-owner w/ BMS	HIF-2a	Renal Cell Carcinoma		Discovery	Discovery	unknown
	SDGR5	Co-owner w/ BMS	SOS1	KRAS-driven Cancers		Discovery	Discovery	Discovery	Preclinical
	-	Co-owner w/ BMS	-	Oncology, Immunology, Neurology			Discovery	unknown
	-	Co-owner w/ BMS	-	Neurology					Discovery
	-	Co-owner w/ BMS	-	immunology
	-	Whole	LRRK2	Neurology				Discovery	Discovery
	-	Whole	PRMT5-MTA	Oncology				Discovery	Discovery
	-	Whole	EFGR(C797S)	Oncology				Discovery	Discovery
	-	Whole	-	Oncology				Discovery	Discovery
	-	Whole	-	Oncology				Discovery	Discovery
	-	Whole	NLRP3	Immunology				Discovery	Discovery
	-	Whole	-	Immunology				Discovery	Discovery
	-	Co-owner w/ BMS	-	Oncology				Discovery	unknown
	-	Co-owner w/ BMS	-	Oncology				Discovery	unknown
	-	Co-owner w/ BMS	-	Immunology				Discovery	unknown
	-	Co-owner w/ BMS	-	Oncology, Immunology, Neurology				Discovery	Discovery
	-	Co-owner w/ Takeda	-	Oncology				Discovery	Discovery
	-	Co-owner w/ Takeda	TYK2	Psoriasis					Phase 2
	-	Co-owner w/ Zai Lab	-	Oncology				Discovery	unknown
	-	Co-owner w/ Lilly	-	Immunology				Discovery	Discovery
	-	Co-owner w/ Ajax	JAK2	Oncology					Discovery
	-	Co-owner w/ Bright Angel Therapeutics	HSP90	Antifungal					Discovery
	-	Co-owner w/ loxo Therapeutics	undisclosed	oncology					Phase 1
	-	Co-owner w/ Morphic Therapeutic	α4β7	IBD					Phase 2
	-	Co-owner w/ Morphic Therapeutic	α4β7	GI indications					Discovery
	-	Co-owner w/ Morphic Therapeutic	αvβ8	Solid tumors, fibrosis					Discovery
	-	Co-owner w/ Morphic Therapeutic	-	Pulmonary arterial hypertension					Discovery
	-	Co-owner w/ NimbusTherapeutic	HPK1	Immuno-oncology					Phase 1/2
	-	Co-owner w/ Otsuka	-	-					Discovery
	-	Co-owner w/ Sanofi	-	oncology					Discovery
	-	Co-owner w/ Structure Therapeutics	APJR	Pulmonary arterial hypertension					Phase 1
	-	Co-owner w/ Structure Therapeutics	-	-					Discovery
	-	Co-owner w/ Structure Therapeutics	LPA1R	Idiopathic pulmonary fibrosis					Preclinical
Insitro	-	Co-owner w/ BMS, TSC alliance	-	Genetic Epilepsies, ALS					Discovery
	-	Whole	-	Solid Tumors					Discovery
	-	Whole	-	MASLD, Obesity					Discovery
Verge Genomics	VRG50635	Whole	PIKfyve	ALS		Discovery	Preclinical	Phase 1	Phase 1/2
	-	Whole	-	Parkinson's Disease				Discovery	Discovery
	-	Whole	-	Parkinson's Disease				Discovery	unknown
	-	Whole	-	Parkinson's Disease				Discovery	unknown
	-	Whole	-	Frontotemporal Dementia				Discovery	unknown
	-	Whole	-	Progressive Supranuclear Palsy				Discovery	unknown
	-	Whole	-	Schizophrenia				Discovery	Discovery
	-	Whole	-	Neurodegenerative Diseases				Discovery	Discovery
	-	Whole	-	Undisclosed				Discovery	unknown
	-	Whole	PIKfyve	COVID-19			Discovery	Preclinical	unknown
	-	-	-	Psoriasis					Discovery
	-	-	-	Atopic Dermititis					Discovery
	-	-	-	Crohn's Disease					Discovery
	-	-	-	Ulcerative Colitis					Discovery
	Partnered Programs	Co-owner w/ Lilly	-	ALS					Discovery
	-	Co-owner w/ Alexion	-	Neurodegenerative Diseases					Discovery
	-	Co-owner w/ Alexion	-	Neuromuscular Diseases					Discovery
Valo Health	OPL-0301	-	S1P1 agonist	Heart failure and Acute Kidney Injury			Phase 1	Phase 2	Phase 2
	OPL-0401	-	ROCK 1/2 inhibitor	Diabetic Retinopathy			Phase 1	Phase 2	Phase 2
	OPAL-0022	-	-	Atherosclerosis			Discovery	unknown
	OPAL-0004	-	-	Atherosclerosis, Giloblastoma			Discovery	unknown
	OPAL-0018	-	-	Atherosclerosis			Discovery	unknown
	OPAL-0003	-	-	Heart Failure, Giloblastoma			Discovery	unknown
	OPL-0101	-	-	Immuno-Oncology		Discovery	Preclinical	unknown
	OPAL-0021	-	-	cancer			Discovery	unknown
	OPAL-0015	-	USP28	NSCLC, Squamous Cell Carcinoma, Targeted Defined Tumors			Discovery	unknown
	OPAL-0024	-	-	Solid Tumors			Discovery	unknown
	OPAL-0001	-	PARP1	Medula/Glioblastoma Brain Tumors, Breast Cancer			Discovery	unknown
	OPAL-0014	-	-	Pancreatic Ductal Adenocarcinoma (PDAC), Targeted Defined Tumors			Discovery	unknown
	OPAL-0023	-	-	Defined Tumors, Immune Modulation			Discovery	unknown
	OPAL-0012	-	USP7	NSCLC			Discovery	unknown
	OPAL-0016	-	-	Induced Neuropathy and Cardiomyopathy			Discovery	unknown
	OPAL-0002	-	-	Neurodegenerative disorders			Discovery	unknown
	OPAL-0006	-	-	Neurodegenerative: Oncology (metastatic)			Discovery	unknown

2023-present moment

In 2023, the first wave of AI-designed (or claimed so) drug candidates hit the breakwater. The industry recorded a number of clinical trial setbacks for some AI-designed drug candidates, including Exscientia's cancer drug candidate EXS21546 which was discontinued out of strategic pipeline prioritisation, as the company representatives explained in an email to BiopharmaTrend.

An AI-inspired schizophrenia drug candidate from partners Sumitomo Pharma and Otsuka Pharmaceutical failed to outperform a placebo in two Phase 3 studies. Sunovion, a subsidiary of Sumitomo Pharma, brought compounds to the alliance that were then screened using PsychoGenics’s SmartCube technology, which employs computer vision to analyze and define behaviors of mice treated with a potential drug.

Adding to negative statistics, BenevolentAI's lead drug BEN-2293 failed to beat a placebo in a Phase 2a atopic dermatitis study, leading to cutting up to 180 jobs and reorganizing its pipeline to conserve cash. The company’s valuation dropped significantly in 2022, has not recovered ever since, and is currently around $145 million (Table 2), a fraction of what it used to be in earlier years.

Notwithstanding years in business and hundreds of millions in raised capital, some AI companies have not produced any clinical drug candidates yet.

For instance, according to a recent Endpoints News interview with Dr. Abraham Heifets, Co-founder and CEO of Atomwise -- the AI in drug discovery pioneer founded in 2012 and raised over $170 million, the company had to cut ‘hundreds of AI discovery programs‘ and focus on just several assets for internal development recently. Currently, all its drug candidates are in preclinical stage.

Founded in 2018, and led by a prominent deep learning scientist Dr. Daphne Koller, Insitro is another example of a company with a seemingly “relaxed” pace of advancing its internal drug candidates. After 6 years in the business, and $643 million of venture capital money raised, the company only has 3 drug candidates in the discovery stage (Table 1).

That being said, Insitro has a strong portfolio of external deals with BMS, Gilead and others, and the company is heavily focused on modeling disease biology. So the limited internal pipeline can be explained by a different focus, and probably we may expect rapid portfolio expansion in the near time.

On the other hand, the industry recorded distinct success stories with growing “AI-inspired” pipelines, like the one of Insilico Medicine, which managed to build an impressive clinical pipeline of 17 preclinical candidates in under 3 years. Several of those are now in clinical trials, including a recent phase 2 candidate for idiopathic pulmonary fibrosis (IPF), five phase 1 candidates for various indications, including kidney fibrosis, inflammatory bowel disease (IBD), immuno-oncology, and COVID-19, and around a dozen preclinical programs in late stages of development.

It is noteworthy that the majority of Insilico Medicine’s AI-discovered programs are first-in-class drug candidates for high-novelty targets or best-in-class candidates for moderately novel targets. The company’s apparent ability to come up with novel targets consistently stems from its target discovery engine PandaOmics.

Another leading player, Recursion Pharmaceuticals seems to be progressing consistently, overall, as the company embraced a robust multifaceted approach to building its end-to-end AI engine Operation System, and its high throughput biology experimentation facilities and robotized lines.

The company has a strong pipeline of two Phase 1 and three Phase 2 candidates, two of which were in-licensed from previous developers. One more candidate did not progress beyond Phase 1 and was put on hold.

It should be noted, however, that according to the Endpoints article, Recursion Pharmaceuticals have not delivered on the promise of ‘generating 100 drugs‘ using AI, as the company advertised a decade ago. It is illustrative of how challenging the drug discovery is, even with the cutting-edge AI technologies and R&D infrastructure.

In 2023 Verge Genomics got positive safety and tolerability data from the Phase 1 clinical trial for its leading candidate VRG50635, a potential best-in-class therapeutic for all forms of ALS. Verge Genomics used CONVERGE™, the company’s all-in-human, AI-powered platform to develop its drug discovery program.

Also, the FDA's clearance of A2A Pharmaceuticals's Investigational New Drug (IND) application for A2A-252, a TACC3 protein-protein interaction (PPI) inhibitor, showcases the potential of AI in accelerating drug development.

Utilizing its AI-driven SCULPT computational platform, A2A Pharmaceuticals, with a lean team of four and limited funding, managed to advance two clinical stage programs, including A2A-252.

AI in Drug Discovery Sector Keeps Consolidating

The AI in the drug discovery and development race has been going for more than a decade now, and today we see all signs of the sector’s consolidation.

Over the last 6 years, there were more than 30 merger and acquisition events in this space. According to the BiopharmaTrend report some of the AI-focused deals in drug discovery, biotech, and clinical trials space included: the 2018 acquisition of Flatiron Health by Roche for $1.9 billion; the 2019 acquisition of Numerate by Valo Health; the 2020 acquisition of Haystack Biosciences by Insitro; the 2021 acquisition of Prescient Design by Genentech; the 2022 acquisitions of Zymergen by Ginkgo Bioworks for $300 million; the 2023 acquisition of InstaDeep by BioNtech for $549 million and double acquisition of Cyclica and Valence Labs by Recursion Pharmaceuticals for around $90 million combined -- to name just a few.

Some of the notable deals are summarized in the figure below:

AI in Drug Discovery Sector Consolidation

Another driver of consolidation is expansion of “big tech” giants into the life sciences, including NVIDIA, Alphabet, Microsoft and others. They leverage their cutting edge resources in artificial intelligence research, world-class tech infrastructures (software, cloud, computational power) and flexible business models.

Famous successes of Alphabet subsidiary DeepMind in protein modeling space, Isomorphic Labs’ recent partnerships with Eli Lilly and Novartis, the advent of NVIDIA’s AI-based Clara Discovery platform for everything from drug design to healthcare research, and Microsoft Cloud for life sciences.

As Dr. Alex Zhavoronkov writes in his Forbes article: “In 2024, I predict NVIDIA will come out big with its healthcare platform so that the need for new AI in drug discovery (AIDD) companies will go away. Drug discovery players will be able to use NVIDIA tools in their cloud, on Amazon, or on massive clusters of NVIDIA GPUs locally.

It is unlikely to hurt established AIDD players with end-to-end AIDD platforms and significant validation, but it will inhibit the formation of startups with little or now differentiation, and it will add more credibility to the field.”

Dr. Zhavoronkov continues about Microsoft: “In 2024, we should expect to see massive growth in LLMs in pharma and biotech, with Microsoft being the main provider. Even when some LLM startups achieve superior performance in benchmarks and come up with new models, they do not stand a chance in big pharma.

Pharma is very conservative - there are too many compliance and legal challenges to deploying generative AI from a new vendor. Microsoft made it easy to use the latest OpenAI models on Azure Cloud and to implement sophisticated AI platform architectures. And Microsoft works all over the world. I predict many internal-external LLM architectures will be built on the Microsoft platform”.

Next, according to the BiopharmaTrend report, venture capital firms are increasingly prioritizing smaller numbers of established AI in drug discovery players over new platform companies, which is a sign of consolidation.

Finally, ‘big pharma’ and leading biotech corporations are increasingly shifting from external partnerships towards building their own AI muscle, effectively diminishing opportunities for novel startups to win R&D dollars.

Big pharma is increasingly choosing well established technology vendors and validated AI for drug discovery engines.

The consolidation of the AI in the drug discovery sector naturally led to formation of a distinct group of AI frontrunners -- companies that managed to build sufficient AI capabilities, robust business models, and win major clients. From the Table 2 below one can see that several companies, including Insilico Medicine, Recursion Pharmaceuticals, Schodinger, and XtalPi demonstrate strong financials, compared to some of the peers in this space:

Disclaimer

Table 2 **

	BenevolentAI ***			Exscientia ***			Insilico			Recursion Pharmaceuticals			Relay Therapeutics			Schrodinger			XtalPi **`
Year	2021	2022	2023	2021	2022	2023	2021	2022	2023	2021	2022	2023	2021	2022	2023	2021	2022	2023	2021	2022	2023Q2
Total Revenue	6,254	12,774	9,334	36,995	32,932	25,565	4,713	30,147	51,180	10,178	39,843	44,575	3,029	1,381	25,546	137,931	180,955	216,666	9,878	19,336	11,299
Drug discovery services							3,687	28,648	47,818							24,695	45,377	57,542	6,189	12,712	5,100
Software solution services							1,026	1,499	3,362							113,236	135,578	159,124	3,689	6,625	6,199
Total Revenue Growth, %		104%	-27%		-11%	-22%		540%	70%		291%	12%		-54%	1750%		31%	20%		112%
Drug discovery services Growth, %								677%	67%								84%	27%		123%
Software solution services Growth, %								46%	124%								20%	17%		95%
Research and development expenses	-76,962	-86,958	-77,380	-59,560	-155,888	-163,535	-38,489	-78,175	-97,341	-135,271	-155,696	-241,226	-172,650	-246,355	-330,018	-90,904	-126,372	-181,766	-33,442	-52,048	-33,124
Operating Loss	-164,052	-238,352	-98,766	-74,230	-175,759	-224,088	-49,359	-83,616		-182,775	-245,727	-350,060	-364,698	-299,275	-373,000	-111,443	-146,817	-177,448	-47,101	-76,171	-61,464
Total current assets	76,567	183,977	116,355	786,694	670,204	515,567	161,541	218,751	188,653	534,718	569,814	438,137	976,242	1,019,505	770,103	625,060	533,989	567,796	616,547	521,516	456,362
IPO Date	2021/12			2021/9						2021/4			2020/7			2020/2
Post-money valuation of last financing round							894.7M												1,968M
Valuation (As of Mar 26th 2024)	93M			749M						2,422M			993M			1,845M
Biological ability	Partnerships			Partnerships			External user base			Partnerships
Chemistry ability	Partnerships			Partnerships			External user base						Partnerships			External user base			External user base
Clinical ability							External user base
Therapeutic Area										Rare & Other, Oncology			Targeted Oncology, Genetic Disease
Phase II	0			1			1			3			1			0			0
Phase I	1			0			5			2			2			2			0
IND enabling	2			2			9			2			1			2			3

** - the data taken from annual reports, see references 26-40

*** - for comparison reasons, values in the UK companies reports (originally in GBP) have been converted to USD at the GBP/USD rate (closing price) as of 31 December of the year depending on the report:
2021 - 1.3522; 2022 - 1.2097; 2023 - 1.2732

**` - for comparison reasons, values in the XtalPi report (originally in RMB) have been converted to USD at the RMB/USD rate (closing price) as of 31 December of the year depending on the report:
2021 - 0.1573; 2022 - 0.145; 2023 - 0.1413

Table 3 ``

Company	Short % of Shares	Current Price (USD) 2024-03-26 (US Trading Hour)	52-Week Price Change %	Current MarketCap (USD'MM)	Last 12 months' Operating Cash Flow (USD'MM)	Last 12 months' Revenue (USD'MM)
Recursion Pharmaceuticals, Inc	18.25%	10.32	64.02%	2,422	-287.8	43.88
Schrodinger, Inc.	7.48%	25.65	5.07%	1,845	-136.7	216.67
Relay Therapeutics, Inc.	10.79%	7.57	-49.80%	993	-300.3	25.55
Exscientia plc	5.42%	5.74	9.19%	749	-149.9	25.5
BenevolentAI	N/A	0.73	-70.25%	93	-72.5	9.27
Insilico Medicine	N/A	N/A	N/A	895	-29.6	51.18

`` - data taken from Yahoo Finance for March 26, 2024.

It is worth noting that XtalPi seems focused exclusively on the contract research business model. At least, the company removed its pipeline information from the english version of its website.

Looking into the future

The AI in the drug discovery space is getting consolidated with the rising barriers of entry for new startups, but also more opportunities for those newcomers that do manage to innovate efficiently, raise money, and leverage novel technologies.

Big tech enters the game

A growing influence of large technology vendors, like Microsoft, Alphabet and NVIDIA can be expected, and they will be increasingly serving as gravitation points for novel startups. For instance, NVIDIA’s Inception program already now nurtures over 1,800 healthcare startups developing cutting-edge, GPU-based tools to optimize operations, enhance diagnostics and develop novel therapeutics.

In 2024, the pharmaceutical and biotechnology sectors are anticipated to significantly expand their adoption of large language models (LLMs), predominantly through Microsoft's Azure Cloud, leveraging OpenAI's models for AI-powered drug discovery (AIDD). Despite the emergence of startups with potentially superior LLMs, the stringent compliance and legal frameworks within the pharma industry favor established providers like Microsoft, which offers comprehensive support for deploying generative AI globally. The trend towards consolidation in the AIDD market is expected to intensify, with a few well-established entities dominating, thanks to their validated platforms and the industry's cautious approach towards adopting innovations from newer vendors.

Era of AI pilots with “big pharma” is over

Big pharma has grown AI muscle to the point that they can oftentimes innovate internally, and depend less on the external pilot programs. The era of numerous AI-deals with everyone around is over. In 2024 and beyond, we may expect an increasing amount of strategic partnerships between big pharma and the most established AI in drug discovery platforms.

Such partnerships will be growing in scope and size of total remuneration.

Already now some deals include not only promise of milestone future payments and royalty payments, but also sizable upfront payments.

In September 2023, Insilico Medicine sold rights to develop and sell a potential cancer drug to California-based Exelixis Inc. for an upfront payment of $80 million and is co-developing another candidate for cancer with Shanghai Fosun Pharmaceutical Group Co., controlled by Chinese billionaire Guo Guangchang.

Around the same time, Exscientia collaborated with Merck KGaA, focusing on AI-driven drug discovery in oncology, neuroinflammation, and immunology, with a $20 million upfront payment and potential milestones up to $674M plus royalties on sales for developed therapies.

AI validation in clinical trials

As more drug candidates from established AI in drug discovery players will be progressing along the clinical trial pipelines, we will have meaningful insights into how AI is really changing the quality of innovation. It is clear that AI can accelerate research and make it cheaper. But does it increase the probability of success? This is yet to be understood in more detail.

At the same time, it is important to mention that while the role of AI in changing the paradigm of early drug discovery is still a matter of validation and more data, the role of AI in drug development is already felt.

In a recent video interview with CNBC, Vas Narasimhan, CEO of Novartis, shares his vision about AI in drug discovery. He thinks it won’t impact early stages of drug discovery within the next five years (we do not quite share this opinion, based on our analysis of several leading companies, such as Insilico Medicine, Recursion Pharmaceuticals, and Schrodinger).

However Vas points out that AI has already impacted drug development and clinical research: e.g.: new trial protocols, working with regulators, working with large patients datasets, etc.

He estimates that AI already helps save 6-9 months of drug development, which can translate into notable financial savings and acceleration.

Data is new priority in the AI race

Finally, the leading AI companies have come to understand the critical importance of generating and possessing their own unique biological data. This realization marks a pivotal shift in the industry's approach to leveraging artificial intelligence for drug development.

It is not enough any longer to have cutting-edge models and algorithms, no matter how sophisticated. It becomes a matter of strategic differentiation to be able to generate or otherwise acquire and control big datasets for model training (e.g. omics, imaging, EHR, etc).

Some companies, like Moderna, Recursion Pharmaceuticals and XtalPi, naturally evolved as digital biotechs with AI-controlled robotized high throughput facilities for running preclinical experiments at scale.

Insilico Medicine is a good example of an AI in drug discovery company that initially evolved as an algorithm-centric company with an efficient collaboration structure. But in the recent couple of years, the company focused massively on augmenting its capability with high throughput experimental capacity of its own, and in 2023, Insilico Medicine launched a 6th generation Intelligent robotics drug discovery laboratory, located in Suzhou BioBAY Industrial Park, China.

Exscientia is yet another example of this trend, as the company opened a facility in Oxfordshire, England, for AI and robotics integration in chemical synthesis and closed-loop drug design, aiming to minimize human intervention and accelerate experimental processes.

Concurrently, the ethical use of AI gains prominence, with international efforts to establish guidelines following the European Union directive and a US executive order, addressing security risks and promoting responsible AI development in the wake of increasing accessibility to generative AI models.

Report methodology

The historical data analysis of therapeutic pipelines (Table 1) has been performed based on the available website snapshots from Web Archive, where we were able to analyze how pipeline diagrams looked in previous periods. In some cases, yearly financial reports were used to get pipeline information for previous years.

The effort was made to track the progression of each molecule or program in any given pipeline over the years, and if the program was unavailable in the next year, it was generally assumed that the program was put on hold for various reasons.

Financial data for Table 2 was obtained from relevant annual reports, in the case of publicly traded companies, and from other publicly available sources for private companies.

Correction policy

In our report on comparing 10 companies, including those that are publicly traded, we made every effort to ensure accuracy and transparency.

Should any factual inaccuracies or outdated information be identified, we will promptly issue corrections in a dedicated section of our report. This correction policy applies to financial data, company profiles, and any comparative analysis presented. Stakeholders are encouraged to report potential inaccuracies to our editorial team via our official contact channels. Corrections will be clearly dated and described, maintaining the integrity of our comparative report and ensuring our readers have access to the most current and accurate information.