Researchers at Harvard University's John A. Paulson School of Engineering and Applied Sciences (SEAS) have developed a novel computational framework that utilizes differentiable programming to understand and engineer cellular self-organization.
The study, published on August 13, 2025, in Nature Computational Science, employs automatic differentiation—a technique crucial for training advanced deep learning models—to predict how alterations in genetic makeup or cellular signaling pathways influence the structure of cell clusters. This approach reframes cell growth as an optimization problem solvable by computational power.
The research team, including graduate student Ramya Deshpande and postdoctoral researcher Francesco Mottes, with senior author Michael Brenner, found that automatic differentiation allows for the efficient calculation of complex functions, thereby pinpointing the precise impact of minor genetic network changes on the collective behavior of cell populations. This innovative method offers a pathway to engineer living tissues with predetermined functions and specific shapes.
This advancement is seen as a significant step towards regenerative medicine and tissue engineering, aiming to repair or replace damaged tissues and organs. The ability to predict and control cellular behavior at a granular level could unlock new possibilities for creating functional tissues and organs. Francesco Mottes noted that this breakthrough offers a promising route to achieving the predictive control necessary for engineering organ growth, a long-sought goal in computational bioengineering.
The findings deepen the understanding of cellular mechanisms underlying biological development and could accelerate the creation of engineered tissues with tailored properties. The computational strategy builds upon advancements in computational modeling, increasingly integrated into regenerative medicine for personalized treatments and optimized tissue engineering approaches.