AI-Powered Virtual Cell Could Become Biology's Universal Simulator

The cell is the fundamental unit of life, a wondrously intricate entity with properties and behaviors that challenge the limits of physical and computational modelling. Despite the challenges, modeling and simulating cell function and behavior is important for understanding how cells work and for determining the root causes of disease.

Now, recent advances in generative artificial intelligence and machine learning, combined with the ability to generate large-scale experimental data, offer ground-breaking opportunities to model cells.

A new landmark paper published in Cell’s 50th Anniversary focus issue on ‘How to Build the Virtual Cell with Artificial Intelligence: Priorities and Opportunities’, sets out a vision for AI-powered Virtual Cells. Forty-two authors from the fields of AI and biology in academia, philanthropy and the private sector, argue that the development of an AI Virtual Cell, an AI-driven computational model that simulates the biological function and interactions of a cell, is possible if the global scientific community tackles the challenge collaboratively across industry and academia. This includes a proposal of their design and how to construct them, the properties and capabilities they should possess as well as how the many parallel efforts in AI for biology and medicine can be integrated into a universal simulator of cell biology.

“Building Virtual Cell Models is no easy feat, but with open science collaborations across biomedicine, a comprehensive predictive understanding of cellular functions and interactions is within reach. So many key leaders have come together to write this Perspective to initiate a global discussion of how we can make this happen,” said Charlotte Bunne, Assistant Professor in Computer Science and Life Sciences at EPFL and lead author of the paper.

The Perspective outlines that two specific revolutions in science and technology – in ‘omics (various disciplines in biology whose names end if the suffix ‘omics’, such as genomics) and in artificial intelligence – have generated an unprecedented opportunity for an ambitious vision of an AI-powered Virtual Cell, a multi-scale, multi-modal large neural network-based model that can represent and simulate the behavior of molecules, cells and tissues across diverse states:

Experimentally, the exponential increase in the throughput of measurement technologies has led to the collection of large and growing reference datasets within, and across, different cell and tissue systems. The amount of data collected has doubled every six months for the past several years. In computer science, advances in AI have enhanced the ability to learn patterns and processes directly from data without needing explicit rules or human annotation. AI satisfies the trifecta of being predictive, generative and queryable, which are key utilities for biological research and understanding.

The paper’s authors argue that by building on these advances, the tools now exist to develop a fully data-driven neural network-based representation of an AI Virtual Cell that is, at some level, agnostic to specific tasks or contexts, and enables new capabilities.

“An AI Virtual Cell should enable a new era of simulation in biology in which, for example, cancer biologists model how specific mutations transition cells from healthy to malignant; developmental biologists can forecast how developmental lineages evolve in response to perturbation in specific progenitor cells; or microbiologists can predict the effects of viral infection on not just the infected cell but also its host organism,” continued Bunne.

“These models will empower experimentalists and theorists alike, transforming the means by which hypotheses are generated and prioritized, and allowing biologists to span a dramatically expanded scope, better fitting the enormous scales of biology and accelerating the discovery of underlying factors behind cellular function,” she said.

By bridging the worlds of computer systems, modern generative AI and AI agents as well as cell biology, the AI Virtual Cell could ultimately enable scientists to understand cells as information processing systems and build virtual depictions of life. As the AI Virtual Cell expands understanding of cellular and molecular systems, it will also increasingly allow researchers to program them and design novel synthetic ones.

The group of AI and biomedical leaders are also staunch advocates for open science and argue that this approach will be crucial to rally community engagement, inspire large-scale international consortia to tackle the challenge and, ultimately, ensure success.

“AI-powered Virtual Cells hold particular potential for personalized medicine, where we are able to generate a patient’s digital twin and align diagnosis and treatment strategies in the context of historically collected measurements across biological systems and studies. They also provide a platform for simulating different choices of treatments and as such support clinical decision making,” said Bunne. “This paper presents a highly ambitious long-term vision of a new era of scientific exploration. With AI Virtual Cells we are optimistic that we can unravel many mysteries of human health and diseases in this century,” she concluded.