The Virtual Biotech

AI for therapeutic discovery

The Virtual
Biotech.

A multi-agent AI framework for therapeutic discovery and development.

An organization of specialized AI scientists that retrieves, analyzes, and integrates biomedical evidence to support research decisions across therapeutic development.

Meet your research team Cite this work
A coordinating AI scientist surrounded by specialists, supporting target prioritization, target validation and modality selection, and clinical translation failure analysis.
Open-source code on GitHub Stanford

Inside the framework

Coordinating evidence
across disciplines.

A virtual Chief Scientific Officer defines the research scope with the user, delegates analyses to specialist agents, and synthesizes their findings. A scientific reviewer evaluates methods and claims, guiding further analysis when evidence is incomplete.

Motivation

Therapeutic development depends on evidence across biological scales, yet the relevant data, analytical tools, and expertise remain fragmented. Integrating these sources and assessing conflicting results is a persistent challenge. The Virtual Biotech organizes specialist analyses within a shared research workflow, with scientific review and traceable evidence to support transparent assessments.

Figure 1 · The Virtual Biotech framework
Three-panel architecture diagram. A: the Chief Scientific Officer and specialist divisions. B: connected data spanning genetics, drugs, trials, diseases, molecular targets, functional genomics, single-cell and tissue expression, pathways, and interactions. C: a question moves through planning, specialist analysis, scientific review, and a final report, with feedback to address gaps.

A Scientific organization
A virtual Chief Scientific Officer coordinates scientific divisions and specialist agents.

B Tools and data
Domain-specific tools query and analyze biological and clinical data.

C Research workflow
The CSO clarifies the question, coordinates analysis and scientific review, then refines the work or synthesizes a report.

From framework to discovery

What the paper explores.

Read in Science

01 / Target prioritization

Learning from
55,984 clinical trials.

A retrospective analysis links the cell-type specificity of drug targets with clinical progression and adverse-event rates.

02 / Target & modality

Exploring B7-H3
in lung cancer.

Genetic, single-cell, spatial, and clinical evidence informs a proposed antibody–drug conjugate strategy, with a focus on the tumor microenvironment.

03 / Clinical translation

Understanding
a trial setback.

An analysis of the vixarelimab trial in ulcerative colitis explores signaling redundancy as a possible explanation for limited efficacy and evaluates a broader gp130-axis biomarker.

The Virtual Biotech supports early-stage therapeutic research. Its analyses generate hypotheses that require further testing and validation.

Cite this work

Harrison G. Zhang et al. The Virtual Biotech: A multi-agent AI framework for therapeutic discovery and development. Science, eaeg6779 (2026). DOI: 10.1126/science.aeg6779.

Interactive workspace

Meet your
research team.

Formulate a question about a target, disease, or therapeutic strategy. Work with the virtual CSO to define the scope, follow specialist analyses, and examine the evidence supporting the conclusions.

For example

“What is the evidence for B7-H3 as a therapeutic target in lung cancer?”