EN
Back to the archive

The archive · Bio & Materials · Technical decision · 2016–2025

Owkin's AI-designed OKN4395 enters the clinic: first patient dosed in Phase I

Owkin bets AI can optimize clinical trials, not just find drugs: OKN4395, its first AI-steered asset, dosed its first patient in Jan 2025 (NCT06789172).

Owkin

The betAI trained on patient data can pick the right asset and optimize the trial around it, beating conventional clinical-development failure rates.Building

What the business is

An end-to-end AI biotech using its K1.0 AI operating system to discover drug targets, select and develop assets, and run AI-optimized clinical trials, while also selling diagnostics.

Starting capitalOver $300M raised, including an $80M Series B-1 from Bristol Myers Squibb in June 2022 (valuation ~$1.5B)

How it started

Founded in 2016 by oncologist Thomas Clozel and mathematician Jean-Philippe Vert, Owkin began as a medical-AI company running drug-trial analytics for pharma, with hospital data networks as the moat. It reached a $1B+ valuation with its June 2022 $80M round led by Bristol Myers Squibb.

What happened

The company licensed OKN4395, a first-in-class EP2/EP4/DP1 triple inhibitor developed from a decade of Idorsia discovery work, then used its K1.0 operating system to pick indications, design the study and set up digital-twin external control arms for early efficacy signals.

How it ended up

On 22 January 2025 the first patient was dosed in INVOKE (NCT06789172), a global Phase Ia/1b first-in-human trial of OKN4395 in advanced solid tumors — the first Owkin asset to reach the clinic under its own AI-optimized program, with combination cohorts alongside pembrolizumab.

Background

Owkin, founded in 2016 by oncologist Thomas Clozel and mathematician Jean-Philippe Vert, started as a medical-AI company running federated analyses on hospital data across Europe and the US. Its bet: models trained on real patient data can improve not just discovery but every step of clinical development.

On 30 January 2025 Owkin announced that the first patient had been dosed on 22 January in INVOKE (NCT06789172), a global Phase Ia/1b first-in-human trial of OKN4395, a first-in-class EP2/EP4/DP1 triple inhibitor for advanced solid tumors. It was the first Owkin program to use the K1.0 AI operating system from asset selection through clinical development.

OKN4395 came from a decade of discovery by Idorsia; Owkin's contribution was choosing it, understanding the mechanism with AI, and designing an optimized trial with biomarker-led patient subtyping and digital-twin external control arms for early efficacy signals. The study evaluates the drug alone and with pembrolizumab.

The milestone is early — Phase I safety data are years from approval — but it converts Owkin's AI thesis into a testable clinical bet: if AI-selected assets and AI-designed trials cut failure rates, Owkin's model beats conventional biotech; if not, the company is just another trial sponsor.

What has to be true

  • Hospital data partnerships gave Owkin training data no US-only AI startup could match, especially in Europe's privacy-restricted setting.
  • Using AI to pick indications and subtype patients attacks the main cause of trial failure: testing in the wrong population.
  • Digital-twin external control arms shrink the control-group cost and time of Phase I.
  • Owning the asset, not just selling software, captures the upside if the trial works.

What can be applied

The deepest AI moat is data you can legally use: Owkin turned hospital partnerships into training data, then trial design — the asset mattered less than the system that chose it.

Aftermath

As of 30 January 2025, Owkin's INVOKE study is enrolling across sites, with Phase Ia dose escalation running and Phase Ib expansion cohorts planned in sarcoma, NSCLC, ovarian cancer and gastric cancer. The company continues to sell diagnostics and research services while advancing OKN4395 and additional AI-selected pipeline assets; the trial's first data will determine whether the AI-optimized thesis holds.

Sources

spotted an error? The archive wants to know.

Your turn

You just read one. Describe what you are building, and see who is betting on the same thing.

Free account · 3 free questions · no card

Related cases