Why Drug Target Selection Is So Expensive
AI drug discovery refers to the use of artificial intelligence systems to read, connect, and interpret huge volumes of biological and clinical data so that scientists can select and design drug candidates faster, with fewer failed experiments and clearer evidence behind each decision. In this context, drug target selection has become one of the most expensive and dangerous choices in pharmaceutical R&D, because it locks companies into a long, costly program before a single patient is treated. A target decision depends on mechanistic biology, feasibility, epidemiology, and market considerations, and weak evidence can send entire pipelines in the wrong direction. When Bayer researchers tried to reproduce the published data behind 67 of their own drug–target projects, the internal results matched only about a quarter of the time, showing how fragile those bets can be.
Inside the Causaly–Microsoft Collaboration
At Microsoft Build 2026, Causaly announced a collaboration with Microsoft aimed at this high‑stakes decision point: which biological target to pursue. Microsoft Discovery produces computational signals from large data assets, while Causaly’s agentic AI platform evaluates whether those signals are worth believing, based on biomedical knowledge and internal data. As Microsoft executive Aseem Datar put it, Causaly is the layer that determines “whether these insights are biologically meaningful and consistent with an organization’s existing use cases and institutional knowledge.” Microsoft supplies enterprise infrastructure, governance, high‑performance computing and, in time, quantum resources. Causaly focuses on the scientific interpretation that turns signal into insight, connecting what is known in external literature with what an organization already knows. Together, they aim to automate the knowledge work around drug target selection, not to replace experimental proof but to make each laboratory bet more informed and less likely to fail.
Knowledge Graphs and Inspectable Reasoning in AI Drug Discovery
Causaly’s agentic AI platform is built on two knowledge graphs that support more transparent pharmaceutical R&D automation. The Bio Graph maps drugs, targets, diseases, and pathways into some 70 million directional relationships, showing what up‑regulates or down‑regulates what, and supports target identification, biomarker discovery, and on‑target safety analysis. The Pipeline Graph captures which company is working on which drug, against which target, and in what phase, combining peer‑reviewed science with press releases and other public signals for competitive intelligence. On top of these graphs, AI agents read external literature and internal datasets, then propose and rank hypotheses about drug target selection. A core design choice is inspectability: scientists can trace each suggested relationship back to the underlying evidence, seeing how many articles support a link and what the original sources say, instead of relying on a black‑box recommendation.
From Data to Insight: Autonomous AI Teams for Science
The Causaly–Microsoft effort points toward autonomous AI teams for scientific research, where specialized agents cooperate across a discovery platform. Microsoft Discovery focuses on finding patterns and signals in enormous biological and chemical datasets, while Causaly interprets those signals through its knowledge graphs and an agentic workflow that encodes mechanistic plausibility, biological rationale, and prioritization. This pipeline produces ranked shortlists of proposed drug targets, biomarkers, or candidates, each with preserved provenance so scientists see why something was recommended. The approach recognizes that biomedicine is a non‑verifiable system for AI: agents can generate five or 500 ideas but cannot confirm them without lab assays or animal studies. Instead, AI aims to compress the reading, synthesis, and protocol‑drafting steps that slow programs, allowing human experts to spend more time on judgment, experiments, and the most promising targets.
What AI Can and Cannot Automate in Drug Discovery
For all the attention on AI-powered biotech, this collaboration draws a clear line between automating knowledge work and running biology itself. A cardiovascular clinical trial still takes months or years; no model can materially speed up the time it takes to see whether a drug works in people. Where AI drug discovery tools help is in building baselines of what is known, designing protocols, reviewing safety histories, and triaging potential targets so that limited experimental capacity is not wasted. Causaly’s platform can propose and rank hypotheses for drug target selection far faster than human teams, and can support trial planning by automating search and synthesis across internal and external sources. Yet hypotheses remain a human responsibility: scientists must decide which targets deserve investment and which should be dropped, keeping a judgment loop at the center of pharmaceutical R&D automation.
