【AI 把新藥探索壓到 9 個月,九個案子有八個還在第一期】💊
麥肯錫整理了 9 個有公開時程的 AI 新藥案子,探索週期從 36 個月到 9 個月,對照組是大藥廠從命中確認到首次人體試驗的中位數 32 到 50 個月。麥肯錫給的結論是快 15% 到 80%,視標的新穎度與藥物型態而定。
九個案子量的不是同一段路。有的從標的確認算到送件申請臨床試驗,有的從命中確認算到首次人體試驗,最快的兩個只算到臨床前候選藥物,而那個階段的大藥廠基準是 16 到 20 個月,不是 32 到 50 個月。麥肯錫在圖下自己標註,這不是同一口徑的比較。
另一行註腳更重要:樣本偏向已經成功、已經公開的案子,中止的幾乎沒有代表。速度看起來快,有一部分是因為慢的那些沒出現在表上。
最硬的資訊在「目前最高臨床階段」那一欄。九個裡面八個停在第一期或第一/二期,唯一進到第三期的是英矽智能的 rentosertib,一款針對特發性肺纖維化的口服小分子藥,9 月 9 日才在北京協和醫院為第一位受試者給藥。
新藥的錢和失敗率從來不在探索階段,在臨床試驗。AI 確實縮短了前段,而那一段本來就是整條路上最便宜的。真正的答案要等第三期讀出數據。
資料來源: 麥肯錫 AI 藥物探索報告、英矽智能公告、美國臨床試驗登記資料庫 (圖片來自 @glocalinvestor)
AI Cut Drug Discovery to Nine Months, and Eight of Nine Are Still in Phase 1
McKinsey compiled nine AI-enabled drug programmes with publicly disclosed timelines. Discovery cycles run from 36 months down to 9, against a large-pharma benchmark of 32 to 50 months from hit identification to first-in-human. McKinsey's conclusion is that these ran 15% to 80% faster depending on target novelty and modality.
The nine are not measuring the same stretch of road. Some run from target identification to a regulatory filing, some from hit identification to first-in-human, and the two fastest stop at a preclinical candidate, a stage whose large-pharma benchmark is 16 to 20 months rather than 32 to 50. McKinsey flags in its own footnote that this is not an apples-to-apples comparison.
A second footnote matters more. The sample is skewed toward successful, well-publicised programmes, with little representation of discontinued work. Part of why the speeds look impressive is that the slow ones are not on the table.
The hardest information sits in the furthest-clinical-phase column. Eight of the nine are in Phase 1 or Phase 1/2. The only one in Phase 3 is Insilico Medicine's rentosertib, an oral small molecule for idiopathic pulmonary fibrosis, which dosed its first patient at Peking Union Medical College Hospital on 9 September.
The money and the failures in drug development have never been in discovery. They are in the clinic. AI has genuinely compressed the front end, and the front end was always the cheap part. The real answer arrives when a Phase 3 reads out.
Source: McKinsey AI drug discovery research, Insilico Medicine announcements and the US clinical trials registry (Image: @glocalinvestor)