Harvard Physicists Develop New Method to Investigate Cellular Self-Organization

Physicists at Harvard have developed a computational method that can uncover the rules that cells use to self-organize, representing it as an optimization challenge that computers can address. This innovative method employs automatic differentiation, originally developed for neural network training, to project how subtle modifications in genes or cellular signals impact the final arrangement of cells. The researchers suggest this predictive capacity could someday enable scientists to design living tissues with precise functions and shapes in actual biological experiments.

Cellular self-organization is a foundational phenomenon in biology, governing how cells naturally form clusters, divide, and transform into complex shapes such as organs, wings, or limbs. Scientists have studied this elaborate behavior to advance synthetic organ creation and better understand cancer. Nonetheless, engineering individual cells to reliably produce targeted group outcomes has long relied on trial and error.

In their study published in Nature Computational Science, the team address this challenge by approaching cellular organization from an optimization perspective. Their computational framework discerns the rules that cells must obey for collective functionality to emerge. These rules are interpreted by the computer as genetic networks, which guide cell behavior by shaping chemical communication between cells and physical interactions that promote adhesion or separation.

This framework currently functions as a proof of concept, but the researchers believe it can be merged with laboratory experiments to enhance the understanding and steering of organismal development at the cellular scale. 

Automatic differentiation, the cornerstone of their methodology, consists of algorithms that precisely calculate the consequences of small changes within gene networks on the behavior of cellular collectives. The team has previously applied these algorithms to other domains such as self-assembling materials, fluid dynamics, and protein engineering. 

First author Ramya Deshpande stated that the principles established in the study could inform future experiments. She asked, “Once you have a model that can predict what happens when you have a certain combination of cells, genes or molecules that interact, can we then invert that model and say, ‘We want these cells to come together and do this particular thing. How do we program them to do that?’” Co-author Francesco Mottes added that with sufficiently accurate predictive models calibrated to experiment, scientists might eventually “engineer the growth of organs” and dictate cellular assembly for desired outcomes.