AI-powered virtual cells will mimic a wide range of cellular behaviors with unprecedented accuracy, scientists say. The cell is an incredibly complex biological factory. The basic unit of life, cells harvest energy from the environment and use it to manufacture complex molecular machines, build copies of themselves, and move. This 4-billion-year-old biological wizardry continues to amaze…
AI-powered virtual cells will mimic a wide range of cellular behaviors with unprecedented accuracy, scientists say.
The cell is an incredibly complex biological factory. The basic unit of life, cells collect energy from the environment and use it to manufacture complex molecular machines, to build copies of themselves and to move. This 4-billion-year-old biological wizardry continues to fascinate biologists who struggle to understand the intricate processes and principles that operate within it.
To change this, scientists have modeled cells at different levels of detail. These models help simulate important biological processes, such as the transcription and translation of proteins. Even the most sophisticated models can predict some general properties of a bacterial cell, such as its phenotype, based on the organism’s genetic code.
But these models are still just a first step toward understanding the full complexity of cellular mechanisms. Perhaps the biggest challenge is simulating the vast number of biological processes and pathways across different scales of resolution, from the atomic and molecular level to the cell and tissue level. These activities are characterized by nonlinear complexity, where small changes in initial conditions can have no effect, or can lead to huge differences in outcomes over timescales ranging from picoseconds to hours or days.
A qualitative leap
Of course, improvements in processing power and algorithmic complexity are making models progressively better. But what scientists need is a paradigm shift that goes beyond these incremental improvements.
Such a shift is now on the horizon, say Charlotte Boni of Stanford University and her colleagues, as artificial intelligence has the ability to model cells with unprecedented precision and at a wide range of scales.
The scientists plan to build an AI-powered virtual cell that can mimic the behavior of real cells with such accuracy that it can predict how cells will respond to a wide range of stimuli, identify potential drug targets, and even evaluate virtual versions of those drugs. The approach, they say, puts “a comprehensive predictive understanding of cell mechanisms and interactions within reach.”
Training data and technological advancements
The key to success with this approach is the availability of training data. This is where bioinformaticians shine. “The volume of raw biological data is undeniable,” Bonnie and colleagues say. They point to the Sequence Read Archive, a public repository of DNA sequence data that currently holds more than 14 petabytes of information, a thousand times more than was used to train the ChatGPT model.
The challenge, of course, is to choose training data wisely from these sources. Much of the existing data may be redundant or of limited value for training purposes, and is not diverse enough to capture the full range of cellular behavior. This is because this data is largely focused on organisms used in laboratory experiments, such as E. coli, mice, and humans, which introduces bias into the resulting models.
Bonnie and colleagues point out that more data are needed from a variety of sources including DNA, RNA, and protein sequences, as well as on sites of activity at the protein, tissue structure, and other levels.
Innovative vision
Another important element is the structure of the model itself. Boni and colleagues propose three levels of interaction that mimic the cell at the molecular, cellular, and multicellular levels. Each level can be interrogated by virtual tools to produce outputs designed to provide insights to humans or as inputs to another virtual tool. In this way, experiments can be designed to evaluate the behavior of cells at different levels.
Currently, the AI-powered virtual cell is still in its early stages, but its ambitions are big. This work cannot be done by one or two research groups alone; it requires extensive collaboration between universities, government, and industry.
Coordinating such collaborations is usually a challenge, but in this case the stakes are high enough to motivate everyone. “The AI virtual cell has the potential to revolutionize the scientific process, leading to future breakthroughs in biomedical research, personalized medicine, drug discovery, cell engineering, and programmable biology,” Boni and colleagues say. We’ll have to see what happens next and who gets involved.
