In January 2020, Exscientia announced that DSP-1181, a molecule designed entirely by its artificial intelligence platform, had entered Phase I human trials in Japan. It was the first time an AI-designed compound had reached that stage. The milestone was widely covered as proof that machine learning could compress the early years of drug research. In May 2023, Sumitomo Dainippon Pharma, Exscientia’s partner on the project, announced it was discontinuing work on DSP-1181 for obsessive-compulsive disorder. The Phase I study had been completed in 2021. The data did not support further work in that indication.
The outcome does not erase what the program demonstrated. Exscientia and Sumitomo Dainippon Pharma took DSP-1181 from target identification to a clinical contender in less than 12 months. The conventional sector timeline for a similar molecule is about 4.5 years. The question the case raises is whether speed alone is enough when the biology is still uncertain.
The Compound and Its Target
DSP-1181 is a long-acting, potent serotonin 5-HT1A receptor agonist. It was developed for obsessive-compulsive disorder, a condition where existing treatments, mostly selective serotonin reuptake inhibitors, leave many patients with inadequate relief. The 5-HT1A receptor has been studied for decades as a target for anxiety and OCD. But no drug had successfully balanced potency, selectivity, and a dosing schedule that patients could tolerate over long periods.
Exscientia’s Centaur Chemist platform was tasked with finding a molecule that hit those criteria simultaneously. The AI system generated candidate structures, predicted their pharmacological properties, and ranked them by the likelihood of success. Chemists then synthesized and tested the top-ranked molecules. The iterative loop between prediction and experiment is what allowed the team to converge on DSP-1181 in under a year.
The Partnership and the Deal
Exscientia, headquartered in Oxford, UK, partnered with Sumitomo Dainippon Pharma, a Japanese pharmaceutical company, to develop DSP-1181. The partnership combined Exscientia’s AI platform with Sumitomo Dainippon’s development infrastructure and its knowledge of the Japanese regulatory environment. The Phase I trial was conducted in Japan under the oversight of the Pharmaceuticals and Medical Devices Agency.
The financial terms of the deal were not disclosed. What is known is that the partnership extended beyond DSP-1181. The two companies continued to work together on other programs, including DSP-2342 for psychosis, which also later failed in Phase I. The broader collaboration structure, in which Exscientia provided AI-designed prospects and Sumitomo Dainippon handled development and commercialization, was typical of the early AI-biopharma partnerships that emerged in the late 2010s.
The Speed Advantage
The headline figure from the DSP-1181 program was the timeline. Exscientia stated that the project took less than 12 months from initial target identification to identifying a clinical contender. The sector average for a similar drug, from target selection to candidate nomination, is approximately 4.5 years. That represents a potential reduction of 75 percent in the discovery phase.
That speed came from the Centaur Chemist platform’s ability to evaluate a far larger chemical space than a human team could. The AI system could screen millions of virtual compounds, predict their binding affinity to the 5-HT1A receptor, and estimate their absorption, distribution, metabolism, excretion, and toxicity profiles. Chemists then focused experimental resources on the most promising prospects rather than synthesizing and testing hundreds of molecules in series.
What the Speed Did Not Change
Speed in discovery does not guarantee speed in development. The Phase I trial of DSP-1181 took more than a year to complete. The data review and the strategic review that followed took another two years. The conventional wisdom that AI can shorten the overall drug development cycle from target to approval remains unproven. DSP-1181 showed that AI can compress the early stage. It also showed that the later stages have their own pace.
Why Development Was Discontinued
Sumitomo Dainippon Pharma said in May 2023 that it was discontinuing work on DSP-1181 for OCD following a strategic review of its pipeline and the results from the Phase I study. The company stated that the data did not support further work in this indication. No specific trial data were released. The decision was a business judgment based on the totality of the evidence, not a single safety signal or efficacy miss.
Discontinuation after Phase I is common in the pharmaceutical sector. The majority of drug prospects that enter Phase I never reach the market. The failure rate for central nervous system drugs is particularly high. OCD is a notoriously difficult indication for drug development because the placebo response in trials is large and variable, and the relationship between receptor occupancy and symptom improvement is not fully understood.
What the Case Means for AI Drug Discovery
The DSP-1181 story is often cited as both a validation and a cautionary tale for AI-driven drug research. The validation is straightforward: an AI platform designed a molecule that met its preclinical criteria and was judged safe enough to test in humans. That was not a given before 2020. The caution is that AI cannot solve the biology problem. A molecule that binds to the right receptor with the right potency and the right half-life still has to work in patients, and that depends on factors the models do not capture.
Exscientia has continued to advance other programs. The company’s pipeline includes prospects in oncology, immunology, and infectious disease. The partnership with Sumitomo Dainippon Pharma continued after DSP-1181 was discontinued, though DSP-2342 also later failed in Phase I. The sector has moved on from the question of whether AI can design a drug to the harder question of whether AI-designed drugs have a higher probability of success than conventionally designed ones. DSP-1181 does not answer that question.
The Bottom Line
Exscientia and Sumitomo Dainippon Pharma achieved a genuine first in January 2020. DSP-1181 was the first AI-designed compound to enter human trials. The speed of the discovery phase, under 12 months versus a sector average of 4.5 years, was remarkable. The Phase I trial was completed in 2021. Development was discontinued in 2023 because the data did not support further work in OCD.
The program did not produce a new drug. It did produce evidence that AI can dramatically accelerate the earliest stage of drug research. Whether that acceleration translates into more approved drugs will depend on how the field integrates AI with a better understanding of disease biology, patient selection, and trial design. As of May 2024, the position since the discontinuation has not been established here.
Key Facts
- Drug compound: DSP-1181
- Therapeutic target: Obsessive-compulsive disorder (OCD)
- Mechanism of action: Long-acting, potent serotonin 5-HT1A receptor agonist
- AI platform used: Centaur Chemist
- Developer: Exscientia (Oxford, UK) in partnership with Sumitomo Dainippon Pharma (Japan)
- Phase I trial start date: January 2020
- Phase I trial location: Japan
- Time from target to candidate: Less than 12 months (sector average: ~4.5 years)
- Phase I completion: 2021
- Development discontinued: May 2023 (data did not support further work in OCD)
- Related programs: DSP-2342 for psychosis (also failed in Phase I)
