AI Accelerates Scientific Discovery, Enhancing Research Capabilities

AI is transforming the landscape of scientific discovery, moving beyond human limitations of time and resources to offer unprecedented computational power and research support. This shift allows scientists to explore complex problems, analyze massive datasets, and validate theories at scales previously unimaginable Source.

AI's role extends to facilitating interdisciplinary collaboration, reviewing extensive literature, generating hypotheses, and interpreting results across various fields, from bioengineering to space exploration.

Advancing Biology with AI

Traditional biological research, often characterized by slow observation and experimentation, is being significantly accelerated by AI. New models enable virtual experiments that can predict gene mutations, design novel genetic sequences, and deepen our understanding of diseases. Stanford researchers are at the forefront of this transformation.

Evo 2: A DNA Language Model

Evo 2, developed by a team including Stanford scholar Brian Hie, is the largest DNA language model ever trained for biology. Launched in February 2025 with an initial Hoffman-Yee Research Grant, this model has 40 billion parameters and was trained on 9 trillion base pairs of genetic data across all life forms, including humans Source.

Similar to a chatbot, Evo 2 takes a string of DNA letters as input and autocompletes the gene sequence. This capability allows scientists to analyze mutated gene sequences, potentially leading to insights into human health and more resilient agricultural crops. According to Hie, "Evo 2 illuminates complex biological processes, such as protein biochemistry, enabling us to accelerate the study of gene function and disease."

Modeling the Human Cell with AI

For centuries, scientists have aimed to model the human cell. AI's capacity to process vast amounts of biological data is bringing this goal within reach. A Stanford team, led by Emma Lundberg, is working on a human-centered foundation model designed to simulate human cells. This could revolutionize drug discovery and personalized medicine.

This virtual cell model would allow experimental drugs to be tested on a digital twin of a patient's cells, predicting efficacies and side effects before administration. The project faces two main challenges:

1. Data Integration: The model needs to recognize diverse biological data types, including DNA/RNA/protein sequences, structures, cellular images, and scientific literature.
2. Multiscale Understanding: It must comprehend how different biological units connect across scales, from molecules to entire organisms.

To facilitate interaction with this complex model, the team developed Biomni, a chat interface for biologists. Lundberg believes that generative models like this are "incredible equalizers" for lowering experimental costs, much like AlphaFold democratized structural biology.

Biomni: AI as a Research Collaborator

Biomedical innovation is often hampered by fragmented research processes, stemming from massive datasets, complex experiments, and numerous analytical tools. Agentic AI, however, offers a solution to scale human expertise.

Stanford, alongside partners from Genentech, Arc Institute, the University of Washington, Princeton, and the University of California, San Francisco, developed Biomni. This biomedical AI agent collaborates with scientists on tasks ranging from gene prioritization and drug repurposing to rare disease diagnosis and microbiome analysis Source.

Biomni integrates hundreds of specialized tools, databases, and software into a unified research environment. With its reasoning capabilities and ability to write Python code, it can autonomously execute complex tasks. Users can prompt Biomni to design wet-lab experiments, support clinical decisions, analyze biomedical data, or review scientific literature. It can even generate new hypotheses without requiring predefined templates.

Jure Leskovec, professor of computer science at Stanford, describes Biomni as "a true collaborator" that "knows biology, is trained on every paper ever published, and can do complex workloads autonomously." To date, 15,000 scientists have used Biomni to automate 100,000 different scientific workflows, making it a widely adopted AI tool in scientific research.

AI's Role in Idea Generation

While AI excels at processing and analyzing data, its ability to generate novel scientific ideas has also been tested. A Stanford study compared ideas generated by human NLP researchers with those from a simple AI ideation agent. Human experts evaluated a mix of human- and AI-generated ideas for novelty, excitement, feasibility, and effectiveness.

The findings indicated that AI-generated ideas were significantly more novel, but often lacked feasibility. Lead author Chenglei Si noted, "In the end, we found that while LLMs have excellent technical creativity, humans come up with more practical proposals, given they are grounded in existing research." While AI's ideas were exciting, the model lacked self-evaluation capabilities and diversity compared to human responses Source.

An additional study considering project execution as part of the assessment further reinforced human superiority when it came to practical outcomes. This suggests that while AI can be a powerful brainstorming tool, human oversight and practical grounding remain crucial for successful scientific implementation.