Artificial intelligence is fundamentally changing how scientists detect disease and conduct research by combining massive datasets with machine learning algorithms. Three major developments show how biobanks, proteomics, and AI agents are converging to make earlier diagnosis possible, speed up laboratory workflows, and identify disease patterns that were previously invisible to researchers (Sources 1, 2, 3).

How Can AI Detect Cancer Earlier From a Simple Blood Test?

Researchers at Barts Cancer Institute, Queen Mary University of London, and the University of Cambridge have developed an AI method called UNITE that detects cancer by analyzing multiple clues hidden in DNA fragments circulating in the blood. The approach works by examining several different signals rather than relying on a single marker, including the length of DNA fragments, how their lengths vary across the genome, the DNA letters at their ends, and changes in copy numbers of different DNA sections.

The challenge is that tumor DNA makes up only a tiny fraction of all cell-free DNA in a blood sample, especially when cancer is at an early stage. By combining multiple signals, UNITE becomes more accurate at recognizing the subtle patterns linked with cancer. When tested on samples the model had never encountered before, UNITE detected cancer in 47% of samples from people with cancer across all stages, including 31% of samples from people with stage I and II cancers. This outperformed two established models for assessing cell-free DNA.

"Each feature lets us look at the data from a different angle. When we combine these signals, the model becomes more accurate and better able to recognize the subtle patterns linked with cancer," said Dr. Haichao Wang, first author of the paper.

Dr. Haichao Wang, Postdoctoral Researcher at Barts Cancer Institute

A key advantage of UNITE is its use of shallow whole-genome sequencing, which requires less data than standard sequencing. This could make the approach cheaper and easier to scale for large populations. The researchers emphasize that liquid biopsies would not replace existing screening methods but could help identify people who need further investigation using imaging or other diagnostic tests.

Why Are Biobanks and Protein Analysis Becoming Central to Precision Medicine?

Biobanks are massive repositories of biological samples paired with detailed patient health records collected over years or decades. These resources are transforming precision medicine by enabling researchers to track disease progression over time and connect molecular data with real-world patient outcomes at scale. The UK Biobank Pharma Proteomics Project, for example, analyzed more than 5,400 proteins across 600,000 samples, making the connection between molecular biology and population health actionable.

Proteomics, the study of all proteins in a cell or tissue, provides a dynamic view of biology that complements the static insights offered by genomics. While genes indicate predisposition to disease, proteins reflect real-time physiological changes happening in the body right now. This dynamic capability makes proteomics particularly valuable for early disease detection.

"When you can characterize thousands of proteins simultaneously, you can see systemic patterns before they surface clinically," explained Dr. Yan Zhang, President of the Proteomic Sciences business at Thermo Fisher Scientific.

Dr. Yan Zhang, President of Proteomic Sciences, Thermo Fisher Scientific

Research from the UK Biobank Pharma Proteomics Project pilot identified protein risk factors for cancer up to seven years before diagnosis appeared. This represents a fundamental shift toward identifying biological signatures of disease before symptoms even emerge. Large-scale collaborative datasets integrating genetic and proteomic data from more than 78,000 individuals have identified over 24,000 protein quantitative trait loci across more than 1,100 proteins, providing new insight into biological mechanisms.

How Are AI Agents Speeding Up Biomedical Research Workflows?

A new AI agent called Biomni, developed by researchers at Stanford University and the Stanford spinout company Phylo, is designed to function as a semi-autonomous research assistant that can execute complex tasks without explicit training examples. The system integrates large language model reasoning and is optimized for "zero-shot learning," meaning it can complete new tasks or classify unknown categories without being trained on examples first.

The bottleneck in modern biomedical research is not data generation but rather converting all that information into robust discoveries. Scientists spend enormous time reading literature, analyzing data, writing code, and designing experimental protocols. Biomni is designed to handle many of these labor-intensive tasks, freeing researchers to focus on higher-level thinking and validation.

"Most AI models in biology today are very specialized for one specific task or data type. They are powerful, but locked into a single problem. Biomni is a general-purpose agent," said Jure Leskovic, Professor of Computer Science at Stanford University and scientific cofounder of Phylo.

Jure Leskovic, Professor of Computer Science, Stanford University

In a study published in Science, the Stanford and Phylo team demonstrated that Biomni can automate labor-intensive laboratory workflows by identifying the correct tools, databases, publications, and types of biology needed to complete a task. For tasks such as finding genes with the strongest effects in a given context or identifying likely causal genes within a genetic locus, Biomni outperformed several other AI systems, including Claude Sonnet, a base large language model.

What Real-World Results Has Biomni Achieved?

When researchers compared Biomni's accuracy and speed to human researchers for tasks related to single-cell RNA-seq annotation, rare disease diagnosis, and genome-wide association study causal gene detection, Biomni's accuracy was roughly equivalent to its human counterparts, although it completed the tasks much faster.

In one real-world test, Biomni autonomously generated and executed a complete analytical pipeline from a prior study to evaluate physiological responses to COVID-19 infection. Using its workflow, Biomni identified biomarkers and calculated effect sizes and correlations that closely matched those of the original study, indicating it can reproduce expert-level analyses.

In another test, Biomni was asked to construct an end-to-end CRISPR cloning protocol involving guide RNA design, cloning, primers for sequencing, and a complete map of the finished plasmid. Researchers took the protocol and ran it exactly as written by Biomni with no changes. The next day, they found colonies and sequenced them using Biomni's primer, finding a perfect match, meaning the cloning was successful.

Steps to Implement AI-Driven Biomedical Research in Your Lab

• Assess Your Data Infrastructure: Evaluate whether your lab has access to well-annotated, longitudinal datasets or biobank samples that can support AI-driven analysis. Large, diverse datasets are essential for training models that generalize across populations and disease types.
• Identify Labor-Intensive Workflows: Document the time-consuming tasks in your research pipeline, such as literature review, data analysis, protocol design, or code writing. These are the areas where AI agents like Biomni can provide the most value by automating routine work.
• Verify Results Independently: While AI agents can speed up research, human validation remains critical. Plan to verify any results generated by AI systems through experimental testing or expert review before drawing conclusions.
• Explore Accessible AI Tools: Phylo offers a free version of Biomni to academic researchers, with discounted paid versions for those needing more computing power. Check whether your institution qualifies for academic pricing or partnerships.
• Collaborate Across Disciplines: AI-driven research benefits from diverse expertise. Work with bioinformaticians, wet-lab researchers, and computational scientists to ensure AI outputs are interpreted correctly and integrated into your experimental design.

What Challenges Remain Before These Technologies Reach Patients?

Despite the promise of these breakthroughs, significant hurdles remain before they become standard clinical tools. The UNITE cancer detection method still misses many cancers, particularly at early stages, and the study did not test its ability to track changes over time. The researchers suggest that approaches like UNITE could eventually complement established screening methods but would not replace imaging, endoscopy, or other diagnostic tests.

For Biomni and similar AI agents, the challenge is ensuring accuracy and preventing hallucinations, where AI systems generate plausible-sounding but incorrect information. Phylo has built in self-correction mechanisms that monitor for common hallucination patterns, but researchers still recommend verifying any answers Biomni delivers.

Population diversity is also critical for ensuring these technologies work across different groups. Proteomic models based predominantly on one population may not translate to another, limiting their utility for broad clinical implementation. Capturing diversity at scale, as in initiatives like Singapore's PRECISE-SG100K program profiling 100,000 volunteers across a uniquely diverse cohort, helps reveal aspects of disease biology that are more broadly meaningful for therapeutic development.

"Population diversity is critical because biological variation is not uniform across populations. A proteomic model based predominantly on one population may not translate to another," cautioned Dr. Yan Zhang.

Dr. Yan Zhang, President of Proteomic Sciences, Thermo Fisher Scientific

The researchers behind these technologies emphasize that AI is designed to augment human expertise, not replace it. As these tools become more common in biomedical science, scientists will need to develop new skills in evaluating AI outputs and interpreting results, but they will remain central to the discovery process. The automation frees researchers from tedious work so they can focus on uniquely human tasks like hypothesis generation, experimental design, and critical thinking.