Translational AI in Biomedicine: From Discovery to Real-World Impact
Organizer: CABS STC | Sponsor: Innovo Health Labs, ChemScene
Overview
On May 30, 2026, the Science & Technology Committee (STC) of the Chinese American Biopharmaceutical Society (CABS) successfully organized the workshop, “Translational AI in Biomedicine: From Discovery to Real-World Impact,” with sponsorship from Innovo Health Labs and ChemScene. The event brought together leading experts from academia and industry to discuss the growing role of artificial intelligence in biomedical research and drug development.
Through presentations and discussions, participants explored how AI-driven technologies can accelerate scientific discovery, streamline translational research, and facilitate the adoption of innovative solutions in real-world healthcare settings. The workshop also examined emerging AI applications across the biomedical innovation pipeline, from early-stage discovery to clinical development, while addressing practical challenges and opportunities in implementation. By fostering interdisciplinary dialogue and collaboration, the event aimed to inspire new ideas and advance the future of AI-enabled biomedical innovation.
The workshop opened with a warm welcome and introduction of CABS by Liping Meng (President-Elect of CABS), followed by an overview of the agenda and speaker introductions by Dr. Alex Yang (Co-Chair of the STC).
Scientific Presentations
The first presentation, “AI agents to accelerate biomedical discoveries,” was delivered by Dr. James Zou (Associate Professor, Stanford University). Prof. Zou outlined that AI "Virtual Labs" can perform complex scientific research. Multiple specialized agents outperform a single general-purpose agent. Groups of agents can develop their own collaborative social dynamics, allowing them to coordinate and solve problems more efficiently.
Prof. Zou also introduced a framework called Paper2Agent, which automatically converts a scientific paper into an AI-powered workflow. AI agents analyze the paper and identify code repositories, datasets, and experimental procedures. AI assistant can then use these tools to apply the paper's methods directly to new datasets. LLMs often fail to correctly use a paper's codebase, datasets, or implementation details. The solution is to use multiple specialized sub-agents to read and interpret the paper, extract relevant resources, build reproducible tools, validate the implementation, and package everything into a robust MCP server that can be reused by AI systems.
The second talk, “Building the life sciences super-entrepreneur with AI engineering commercialization back to discovery,” was presented by Dr. Xi Fang (CEO, Innovo Health Labs). Dr. Fang focused on how AI can transform biotech entrepreneurship by closing the gap between scientific discovery and commercialization strategy. The speaker argued that many biotech startups fail or receive lower valuations not because of weak science, but because founders lack access to critical commercial intelligence such as regulatory pathways, reimbursement strategy, competitive landscape analysis, IP positioning, and deal benchmarking. Traditionally, obtaining this information required expensive consultants, senior advisors, and months of work, creating a major structural disadvantage for scientific founders.
He introduced the concept of the “AI-augmented founder,” where AI systems are used to dramatically increase research and commercial analysis throughput while human scientists retain responsibility for scientific judgment, experimental design, clinical decisions, and strategic leadership. The proposed platform emphasizes cited sources, reproducibility, auditability, and regulatory-grade evidence tracking to support trustworthy decision-making in life sciences. Overall, Dr. Fang highlighted a future in which small scientific teams, supported by AI infrastructure and expert advisors, can operate with capabilities that previously required large commercial organizations.
The 2nd half session after break continued the workshop’s discussion on how genome engineering, artificial intelligence, and automation are transforming biomedical research and therapeutic development. The two talks focused on two closely connected themes: building next-generation molecular tools and AI co-scientists for genome and cell engineering, and applying AI across drug discovery and development.
From Code to Cure: Building Molecular Tools and AI Co-Scientists to Drive Genome and Cell Engineering
Dr. Le Cong (Associate Professor, Stanford University) presented their recent work with collaborators on developing molecular tools and AI systems to accelerate genome and cell engineering. He began by highlighting the rapid progress of modern biomedicine: genome sequencing has moved from a years-long process to approximately one day for a single human genome, while genome engineering has evolved from individual edits to large-scale perturbation screens involving millions of changes. CRISPR has become a standard tool in biology because it is powerful, useful, and reliable, but the speakers emphasized that biological systems remain highly complex and require new tools to better understand genome evolution, function, and disease mechanisms.
One major focus was the challenge of safe and scalable genome writing. While Cas9-based approaches are highly effective for many applications, they can struggle with efficient DNA insertion and knock-in editing. The speakers introduced work on phage-derived single-strand annealing proteins, or SSAPs, which are compact proteins capable of supporting kilobase-scale gene editing. After screening approximately 600 SSAPs, the team used AI-enabled optimization to improve gene-insertion efficiency. This effort led to EvoSSAP variants with substantially higher activity and demonstrated that machine learning models can identify functional sequence features beyond those obvious from natural evolution.
The presentation also showed how structural mapping helped explain activity-enhancing mutations. These findings suggest that AI-guided protein engineering can reveal unexpected design principles and help expand the development of improved molecular editors.
The speakers then discussed the role of AI in accelerating scientific reasoning. Prof. Cong introduced Genome-Bench, built from more than a decade of real scientific notes and discussions to assess large language model reasoning in genomics. He also described how reinforcement learning can improve AI reasoning for genomic tasks while reducing the need for very large models, potentially lowering the cost of AI-assisted research.
Another highlight was CRISPR-GPT, a multi-agent AI scientist designed to automate genome engineering research. The system integrates scientific reasoning, internal laboratory knowledge, and task-specific agents to support experimental design and decision-making. The speaker shared multiple application examples, including studies related to lung cancer metastasis, APOE4 editing for Alzheimer’s disease, and AI-assisted drug target screening. In one case, a CRISPR-GPT-Screen agent prioritized NK immunotherapy targets and reduced a task that would typically take human researchers several days to only a few minutes, followed by wet-lab validation.
Dr. Yingcheng Wu, a postdoc from Dr. Cong’s lab, further discussed the gap between the speed of AI reasoning and the practical realities of wet-lab research. While AI systems can rapidly generate hypotheses and experimental plans, physical experiments still require careful execution, quality control, and reproducibility. To bridge this gap, the team introduced LabOS, a self-improving AI system for closed-loop dry- and wet-lab reasoning and action, developed with collaborators including NVIDIA and Viture. Dr. Wu also presented LabSuperVision, a specialized vision-language model benchmark for laboratory tasks. Together, these efforts aim to make lab research more AI-operable, AI-reproducible, and ultimately more scalable.
The Future Is Nearer Than You Think: Applying AI for Drug Discovery and Development
Dr. Michelle Chen (CEO, Form Bio), then presented how AI is reshaping drug discovery and development. Her talk emphasized that traditional drug discovery remains slow, costly, and inefficient, with thousands of rare diseases still lacking approved treatments.
Dr. Chen introduced a broad AI toolkit that is changing the field. She described AI as a layered system in which specialized agents perform specific functions, while agentic AI can coordinate these agents to complete complex tasks with limited human input. This agent-based model has the potential to support many stages of drug development, from target discovery and molecule design to data analysis and development decision-making.
Through case studies, Dr. Chen discussed the promise and limitations of end-to-end AI drug discovery. Drawing on examples from Insilico Medicine and other frontier efforts, she highlighted how AI platforms can accelerate the identification of therapeutic targets and drug candidates. However, she emphasized several important lessons: data quality is more important than model complexity; collaboration between biologists and machine learning engineers is essential; interpretability matters deeply in pharmaceutical development; and speed should not mean cutting corners.
Dr. Chen also discussed Form Bio’s work applying AI to genetic medicine and therapeutic development. She highlighted the importance of using deep learning and computational platforms not only to generate ideas, but also to improve prediction, design, analysis, and development workflows. Her presentation reinforced that AI will be most impactful when it is integrated into real scientific and translational processes.
Panel Discussion
The afternoon session concluded with a panel discussion moderated by Dr. Lin Wang (Co-chair of the STC) on the opportunities and bottlenecks in translating AI-enabled innovation into real-world biomedical impact. Panelists identified talent as a major limiting factor, especially the need for people who can connect biology, machine learning, engineering, and commercialization. They also emphasized that AI must be grounded in reality through reproducible raw data, strong validation, and the integration of public information with proprietary know-how.
The panelists discussed how organizations can adopt AI more effectively. They noted that successful adoption requires more conversations across different areas of expertise so teams can better understand what AI can and cannot do. Rather than simply adding AI tools, organizations need to define practical use cases, understand current technical boundaries, and build workflows that combine AI output with human expertise.
When asked which tasks AI agents may perform independently in the next three to five years, panelists expressed cautious optimism. AI may increasingly support literature review, experimental planning, target prioritization, screening design, and data analysis. However, the panelists were skeptical that AI will replace human intelligence, experience, or judgment in the near future. They emphasized the need to distinguish between AI-generated content that requires human validation and AI-generated insights that may be original and useful but still require human decision-making.
The discussion also addressed the role of humans in an increasingly automated research environment. Panelists agreed that AI can help automate and scale many tasks, but human oversight will remain essential for verification, auditing, interpretation, and final decision-making. In commercialization especially, AI can accelerate the process, but humans still need to evaluate risks, make strategic choices, and take responsibility for decisions.
Audience questions further explored what skills junior scientists should develop in the age of AI-powered robotics and lab automation. Panelists encouraged young researchers to focus on problem definition, scientific judgment, and strategic thinking. As AI agents take over more routine tasks, scientists will need to become better at identifying the right problem, choosing the most efficient path forward, and understanding what unique contribution will move a project toward its goal. One optimistic view shared during the discussion was that AI agents may enable more scientists to work with PI-level independence and productivity.
The panel also discussed how small biotech companies can compete with large pharmaceutical companies in the AI era. Panelists emphasized that small companies must remain disciplined, particularly before clinical approval. Survival depends on focus, careful capital allocation, and prioritizing the experiments and decisions that create the greatest value. AI can help small teams move faster and compete more effectively, but it does not replace the need for strategic clarity and scientific rigor.
Closing remarks
Overall, the workshop highlighted both the excitement and the responsibility surrounding AI-enabled research and development. The workshop concluded with strong enthusiasm and overwhelmingly positive feedback. A networking lunch provided a relaxed environment for continued dialogue, fostering new connections and potential collaborations.
This workshop was made possible through the generous support from Innovo Health Labs and ChemScene.
