Discovery

GenBio AI unveils multiscale virtual cell model spanning DNA to phenotype

A Palo Alto-based AI company has released what it describes as a multiscale, stateful simulation of the human cell, an approach that differs architecturally...

GenBio AI unveils multiscale virtual cell model spanning DNA to phenotype

Palo Alto-based GenBio AI has unveiled AIDO Cell, a virtual cell model designed to simulate biological responses across multiple scales, from DNA and RNA through proteins to whole-cell behavior. The company is positioning the system as a step beyond cellular foundation models centered primarily on transcriptomic data, with an architecture that can model cascading effects across biological layers and retain the cumulative impact of sequential perturbations.

AIDO Cell's key distinction, according to GenBio, is that it is both multiscale and stateful. Earlier cellular foundation models such as Geneformer and scGPT have largely focused on representations derived from single-cell transcriptomic data. GenBio said AIDO Cell instead links DNA, RNA, protein, and cellular representations within a single simulation, allowing a perturbation at one level — such as a genetic alteration or drug treatment — to propagate through the modeled biological hierarchy. The system also retains changes caused by previous perturbations, enabling researchers to model sequences of interventions rather than treating each independently.

As an initial proof-of-concept, GenBio said AIDO Cell reproduced the known mechanism of imatinib in K562 leukemia cells, tracing the effects of treatment across its modeled cellular environment. The company did not disclose quantitative performance metrics, comparisons with competing models, statistical analyses, or prospective experimental validation, making it difficult to assess how accurately the system predicts previously unseen biological responses. AIDO Cell currently supports K562 and HepG2 cell lines, with additional cell types in development.

The new system builds on GenBio's earlier AIDO.Cell-100M, a 100-million-parameter transcriptomic foundation model trained on approximately 50 million cells and available through the Chan Zuckerberg Initiative's Virtual Cells Platform. AIDO Cell extends that work toward an integrated simulation spanning multiple biological scales rather than operating primarily on transcriptomic representations.

GenBio was founded by a group of researchers spanning artificial intelligence, computational biology, and protein science. Its co-founders include 2024 Nobel chemistry laureate David Baker, director of the Institute for Protein Design at the University of Washington; Eric Xing, president and university professor at Mohamed Bin Zayed University of Artificial Intelligence (MBZUAI) and professor at Carnegie Mellon University; Carnegie Mellon computational biologist Ziv Bar-Joseph; Stanford and KTH Royal Institute of Technology researcher Emma Lundberg; and former BioMap chief technology officer Le Song. Fred Hu, founder of Primavera Capital Group, serves as GenBio's CEO.

The AllSci BriefSystematic R&D and deal news. Daily.

The release comes amid growing investment in AI-based virtual cell models intended to predict how biological systems respond to genetic and pharmacological perturbations. Xaira Therapeutics unveiled X-Cell in March 2026, a diffusion language model trained on 25.6 million experimentally perturbed cells across seven CRISPR interference Perturb-seq screens. While X-Cell is designed primarily to predict transcriptional responses to genetic perturbations across cellular contexts, GenBio is making a broader architectural claim for AIDO Cell: that perturbations can be propagated across multiple molecular and cellular scales while preserving the effects of previous interventions.

The field remains at an early stage, with independent studies raising questions over whether large cellular foundation models consistently outperform simpler approaches on tasks including perturbation prediction. For AIDO Cell, the critical test will therefore be whether its multiscale architecture can prospectively predict previously unseen cellular responses and outperform existing approaches when those predictions are subjected to experimental validation.

GenBio said it plans to release more advanced versions of AIDO Cell with expanded capabilities through 2026 and 2027 and is preparing an early-access program for academic, biotechnology, and pharmaceutical researchers.


Access the AllSci platform to explore the science behind the news.


Spot something wrong? Report an issue with this article