Open experimental data may commoditize benchmark performance while increasing the value of proprietary execution, feedback loops and rights to compounds.
OpenADMET's new $34.8 million funding package looks like an AI-for-science grant story. Its more consequential meaning is about industry structure: absorption, distribution, metabolism, excretion and toxicity data are being financed as public infrastructure rather than rebuilt behind every company's firewall.
According to Octant's September 30 announcement, the package includes $15 million from the OpenAI Foundation, $4.9 million from the Gates Foundation and $14.9 million in ARPA-H Phase II funding. The consortium plans to generate prospective experimental data, release datasets and models, and run blind prediction challenges. ARPA-H's public record separately lists UCSF as the prime awardee for the underlying AVOID-OME project, with an award of up to $30.5 million beginning in 2024. [Octant; ARPA-H]
当实验数据和盲测基准成为公共基础设施，模型成绩更容易比较，真正稀缺的价值会转向实验闭环、专有数据和资产权利。
OpenADMET 新获得的3480万美元资助，看上去是一则 AI4Science 融资新闻。更深一层，它在改变行业的成本结构：药物吸收、分布、代谢、排泄和毒性数据，开始被当作公共基础设施投入，而不再由每家公司关起门来重复建设。
根据 Octant 9月30日的公告，这笔资金包括 OpenAI Foundation 的1500万美元、盖茨基金会的490万美元，以及 ARPA-H 的1490万美元二期资金。项目计划生成前瞻性实验数据，公开数据集与模型，并用盲测挑战检验预测能力。ARPA-H 的公开记录则单独确认，底层 AVOID-OME 项目自2024年启动，由 UCSF 牵头，整体资助上限为3050万美元。[Octant；ARPA-H]