神經科學領域長期面臨藥物研發困境，Alzheimer's 與 Parkinson's 等疾病的病因至今尚未完全闡明，致使製藥業者反覆針對有限的已知生物靶點進行開發，而絕大多數候選藥物最終均以失敗告終。新一代人工智慧公司正試圖從根本上改變這一局面：不再僅僅利用 AI 針對已知靶點設計更優藥物，而是透過 AI 發現全新的生物學機制，從而揭示大量未經探索的靶點。Recursion Pharmaceuticals 執行長 Najat Khan 指出：「建立新穎的數據集並從中發現靶點，這一路徑確實可行。」該公司正利用其專有的大規模數據集來挖掘先前未被探索的神經科學靶點。
瑞士製藥巨頭 Roche 及其美國生物技術子公司 Genentech 已率先對這一理念投下業界最早的重大賭注。雙方與總部位於 Utah 的 Recursion 合作推進了首個項目——該生物技術公司利用 AI 為一種未公開的神經科學疾病識別出一個此前從未被探索的生物靶點。這一新靶點的發現，源於 AI 對超過一萬億個（over 1 trillion）實驗室培養的人類神經元數據的分析，並經過了實驗室的廣泛驗證。Khan 表示：「這只是眾多成果中的第一個。」
更大的價值並非來自單一靶點，而是其背後的數據集本身。Recursion 認為，與傳統「實驗完成即棄置數據」的模式不同，該公司已構建了一個可持續挖掘的資源，無需每次從零開始收集與重建數據，從而顯著加速研發進程。公司神經科學副總裁 Christopher Winrow 將其比喻為一張「地圖」，研究人員可以「不斷回到這張地圖上」搜索更多藥物靶點。儘管這一新靶點仍有可能失敗——大多數實驗性藥物確實如此——但關鍵問題在於此流程能否被重複驗證。若 AI 能持續揭示科學家此前未曾關注的生物學機制，將有望從根本上改變藥物發現的起點。
The neuroscience field has long struggled with drug development, as the causes of diseases such as Alzheimer's and Parkinson's remain incompletely understood, forcing drugmakers to repeatedly target a limited shortlist of known biological targets—most of which ultimately fail. A new generation of AI companies is attempting to fundamentally change this paradigm: rather than using AI solely to design better drugs against known targets, they aim to discover entirely new biology and uncover a wealth of unexplored targets. Recursion Pharmaceuticals CEO Najat Khan states that "creating novel data sets and then finding targets can actually work," as her company leverages proprietary large-scale datasets to identify previously unexplored neuroscience targets.
Swiss pharmaceutical giant Roche and its US biotechnology arm Genentech have placed one of the industry's first major bets on this approach. They advanced the first project from their partnership with Utah-based Recursion, after the biotech used AI to identify a previously unexplored biological target for an undisclosed neuroscience disease. This new target emerged from AI analysis of data derived from over one trillion laboratory-grown human neurons and was subsequently validated extensively in the lab. "This is the first of many," Khan says. (Key numbers: 1)
The greater promise lies not in any single target but in the dataset itself. Recursion believes it has built a continuously mineable resource rather than following the traditional model where data is discarded after each experiment, thereby significantly accelerating the speed of development. Christopher Winrow, the company's vice president of neuroscience, likens it to a "map" that researchers can "continue to go back to" in search of additional drug targets. While this new target may still fail—as most experimental medicines do—the critical question is whether this process can be reliably repeated. If AI can consistently uncover biology that scientists were not previously examining, it could fundamentally change where drug discovery begins.