The Era of AI-Driven Pharmaceutical Industrialization Has Arrived: Entering the Phase of Scalable Clinical Value Realization

(Source: Shangguan News)

XtalPi's On-Premises AI Autonomous Laboratory

“Future new drug R&D will increasingly resemble a systems engineering endeavor. We aim to transform many issues originally defined as scientific questions and challenges into engineering problems and challenges.” At the AI-Driven New Drug R&D Innovation Summit held recently in Zhangjiang, Dr. Ma Jian, Co-founder and CEO of XtalPi, stated that the industrial era of AI-driven drug discovery has arrived. AI is evolving from a mere R&D tool into new foundational infrastructure for research and development, advancing from “point breakthroughs” to “full-chain innovation.”

Currently, the consensus in the AI-driven drug discovery industry is shifting toward the realization of clinical value. How to leverage artificial intelligence,RobotCutting-edge technologies that shorten R&D cycles and reduce the risk of clinical failure have become core imperatives for the industry. At the event, XtalPi, together with its partners, unveiled a series of latest clinical and technological advancements, further confirming that the engineering capabilities of AI-driven drug discovery are entering a phase of large-scale realization.

Transforming Scientific Uncertainty into Engineering Certainty

In recent years, the pervasive integration of AI across the entire new drug development pipeline has been reshaping the global pharmaceutical industry. Worldwide, multinational pharmaceutical companies are significantly deepening the penetration of AI within their R&D pipelines; Eli Lilly has publicly disclosed that AI is now extensively integrated into its research and development processes. Meanwhile, the U.S. Food and Drug Administration (FDA) is systematically promoting the application of AI technologies from a regulatory perspective, shifting its stance on novel methodologies such as organ-on-a-chip from merely “supporting scientific research” to “proactively incorporating them into the regulatory framework.” Isomorphic Labs, spun out of Google DeepMind, entered into strategic partnerships worth nearly $3 billion with Eli Lilly and Novartis in 2024, establishing multiple AI-driven drug development pipelines targeting complex modalities including GPCRs, with the goal of advancing these candidates into clinical trials by 2026.

Regarding the future trends of AI-driven drug discovery, Ma Jian provided an assessment from three perspectives: the development of vertical models, the production of R&D data, and the penetration of foundation models.

“In recent years, vertical AI models for drug discovery have been evolving from foundational physics-based models combined with machine learning algorithms toward all-atom models and cross-modal approaches. Previously, there were clear distinctions between algorithmic models for small molecules, large molecules, peptides, and small nucleic acids; however, algorithms and underlying infrastructure will increasingly converge in the future,” stated Ma Jian. “At every stage of drug R&D, data remains highly incomplete—whether in terms of volume or quality, sample space distribution, or uncertainty in objective functions—thus continuing to rely heavily on new experimental infrastructure. We must reexamine the transformation of data generation methods at the R&D stage: whereas human effort previously dominated, automation, robotics, and AI agents will increasingly join the process, collectively building a data flywheel and establishing more intelligent closed-loop wet-dry R&D capabilities.”

Regarding the penetration of foundation models, he believes that with the rapid iteration and development of large foundation models, which were at an elementary level not long ago but have now reached a “Ph.D. graduation” stage, they may attain industry-expert proficiency within a few more months. Consequently, these models will gradually permeate more processes and domains. For instance, in process-oriented organizations, the integration of large foundation models with AI agents can replace many decision-making steps, enabling greater automation and efficiency.

For innovative R&D organizations, how foundational large models can foster closer collaboration with teams in drug development, and how to shift the role of these models from empowering individuals—such as by generating code and documents—to empowering entire organizations to drive collaborative innovation, remains an unresolved challenge in the industry.

Ma Jian further stated that we are increasingly recognizing that the future leaders in this field will no longer be those with the strongest capabilities in any single area, but rather platform-driven forces capable of better orchestrating an ecosystem that integrates technology, science, clinical practice, and industry. Future new drug development will undoubtedly be a systematic engineering endeavor, transforming greater scientific uncertainty into engineering certainty, thereby achieving large-scale industrial implementation and real-world validation.

From Innovation “Nurseries” to Industrial “Forests”

In this wave of global industrialization of AI-driven drug discovery, XtalPi, as a pioneer in global AI-based drug R&D, has taken the lead in establishing an industry-exclusive closed-loop R&D system: “AI model prediction – robotic execution of wet-lab experiments – data feedback to AI – Multi-Agent intelligent scheduling.”

XtalPi’s Chief Scientific Officer, Dr. Zhang Peiyu, provided insights into the transformation of R&D paradigms and AI-driven autonomous discovery systems. He stated that drug development is moving away from empirical and theoretical science, entering an era of data-driven autonomous discovery. Autonomous discovery involves AI independently generating new targets, mechanisms, or compounds, and determining viable candidates through evaluation to establish a closed-loop system for autonomous experimentation.

Taking chemical scenarios as an example, he introduced the workflow of the autonomous discovery system: through natural language descriptions, agents generate R&D workflows and issue instructions; physical intelligence then performs experimental operations, while agents invoke tools to advance experiments and analyze results. This system has been deployed and validated internally at XtalPi, resulting in significant improvements in molecular synthesis efficiency and success rates.

Zhang Peiyu stated that the laboratory serves as an ideal training ground for physical intelligence. Compared to industrial and domestic robotic scenarios, laboratories offer controllable layouts, limited operational objects, and high-value outcomes (such as the creation of new substances, drugs, and materials). Their non-fixed workflows and high scheduling complexity support multi-stage technological iterations of physical intelligence. XtalPi completed the establishment of its AI-driven robotic laboratory capabilities in 2020, officially launched the research and development of general-purpose physical intelligence in 2025, and leveraged its self-built integrated scientific research data infrastructure to achieve a closed-loop process for data generation, annotation, and modeling, thereby continuously improving the overall success rate of new drug development.

In the future, XtalPi will continue to connect various industry stakeholders through an open ecosystem, leveraging its constantly evolving AI and robotics R&D infrastructure to address more pain points in new drug development, thereby advancing Chinese AI-driven drug innovation from technological breakthroughs to tangible clinical value. The foundation and future of this endeavor rest on the continuous building and deepening of engineering capabilities in AI-powered drug research—a core competency of XtalPi and a key factor enabling the entire industry to evolve from an innovation “nursery” into a thriving industrial “forest.”

Original Title: "The Era of Industrialized AI Drug Discovery Has Arrived! Entering the Phase of Large-Scale Realization"

Column Editor-in-Chief: Rong Bing

Source: Author: Wen Hui Bao, Tang Weijie