Artificial intelligence regenerates planaria
For the first time, artificial intelligence proves it can do more than just crunch numbers — and shows us the body plan planaria follow to regenerate.
For the first time, an artificial intelligence system has reverse-engineered regeneration in planaria, small worms whose extraordinary power to regrow parts of their body — you can split their head down the middle and a planaria will grow TWO heads — has made planaria a classic research model for studying the possibilities of human regenerative medicine.
The discovery by Tufts University biologists presents the first model for regeneration discovered by computer scientists using artificial intelligence, and the first complete model of planaria regeneration.
How regeneration works in planaria has eluded human scientists for over 100 years. Published June 4, 2015 in PLOS Computational Biology, this new work demonstrates how "robot science" can help human scientists create the future.
In order to bioengineer complex organs, scientists need to understand mechanisms by which shapes are produced in a living organism. However, a significant gap persists between molecular genetics and understanding how and why a particular complex shape is generated to the correct size and orientation.
"Most regenerative models today that are derived from genetic experiments are arrow diagrams, showing which gene regulates which other gene. That's fine, but it doesn't tell you what the ultimate shape will become. You can't tell if the outcome of many genetic pathway models will end up looking like a tree, an octopus or a human.
"Most models show some necessary components for the process to happen, but not what dynamics produce the shape, step by step. What we need are algorithmic or constructive models, which could be followed precisely so there would be no mystery or uncertainty. Follow the recipe - out comes the shape."
Michael Levin PhD, senior author, Vannevar Bush professor of biology, Director of the Tufts Center for Regenerative and Developmental Biology.
We need to follow models of expected outcomes in order to know where and what triggers are needed to cause regeneration of a particular organ or tissue or shape. However, no such tools exist to sift through the fast-growing mountain of experimental data in regenerative and developmental biology, according to Daniel Lobo PhD, post-doctoral fellow in the Levin lab and the paper's first author.
To address this lack of a model, Lobo and Levin developed an algorithm to produce regulatory networks which can "evolve" and predict results of laboratory experiments.
"Our goal was to identify a regulatory network that could be executed in every cell in a virtual worm so that the head-tail patterning outcomes of simulated experiments would match published data."
Daniel Lobo, post-doctoral candidate, Levin laboratory.
Initially, random regulatory networks could not produce any results. New networks were were then created by combining previous work and performing random changes, additions and deletions. Each network was then tested in a virtual, simulated worm. The algorithm compared all the resulting shapes from simulations with real data existing in databases. Evolution of the algorithm proceeded, gradually, until new networks explained more uniformly how head versus tail regenerated.
Ultimately researchers applied the algorithm to a dataset of 16 key planaria experiments to see if it could identify the regulatory network needed for re- generation. After 42 hours, the algorithm correctly predicted all 16 experimental outcomes in the data. A network of seven known regulatory molecules plus two proteins were identified that had not existed in papers on planarian regeneration before.
"This represents the most comprehensive model of planarian regeneration found to date. It is the only known model that mechanistically explains planaria head-tail polarity under many different functional experiments. It is the first regenerative model discovered by artificial intelligence."
Michael Levin PhD
Lobo and Levin are both trained in computer science, bringing an unusual perspective to developmental biology. Levin majored in computer science and biology at Tufts before earning his Ph.D. in genetics. Lobo earned his computer science PhD before joining the Levin lab.
Their paper represents a successful application of the growing field of "robot science." Levin feels computer science can greatly help basic research by quickly crunching enormous datasets, and with the right algorith, find patterns.
"While the artificial intelligence in this project did have to do a whole lot of computations, the outcome is a theory of what the worm is doing. Coming up with theories of what's going on in nature is pretty much the most creative, intuitive aspect of a scientist's job.
"One of the most remarkable aspects of the project was that the model we found was not a hopelessly-tangled network no human could understand, but a reasonably simple model that people can easily comprehend. All this suggests to me that artificial intelligence can help with every aspect of science, not only data mining but also infering meaning from data."
Michael Levin PhD
A mechanistic understanding of robust self-assembly and repair capabilities of complex systems would have enormous implications for basic evolutionary developmental biology as well as for transformative applications in regenerative biomedicine and the engineering of highly fault-tolerant cybernetic systems. Molecular biologists are working to identify the pathways underlying the remarkable regenerative abilities of model species that perfectly regenerate limbs, brains, and other complex body parts. However, a profound disconnect remains between the deluge of high-resolution genetic and protein data on pathways required for regeneration, and the desired spatial, algorithmic models that show how self-monitoring and growth control arise from the synthesis of cellular activities. This barrier to progress in the understanding of morphogenetic controls may be breached by powerful techniques from the computational sciences—using non-traditional modeling approaches to reverse-engineer systems such as planaria: flatworms with a complex bodyplan and nervous system that are able to regenerate any body part after traumatic injury. Currently, the involvement of experts from outside of molecular genetics is hampered by the specialist literature of molecular developmental biology: impactful collaborations across such different fields require that review literature be available that presents the key functional capabilities of important biological model systems while abstracting away from the often irrelevant and confusing details of specific genes and proteins. To facilitate modeling efforts by computer scientists, physicists, engineers, and mathematicians, we present a different kind of review of planarian regeneration. Focusing on the main patterning properties of this system, we review what is known about the signal exchanges that occur during regenerative repair in planaria and the cellular mechanisms that are thought to underlie them. By establishing an engineering-like style for reviews of the molecular developmental biology of biomedically important model systems, significant fresh insights and quantitative computational models will be developed by new collaborations between biology and the information sciences.
This work was supported with funding from the National Science Foundation grant EF-1124651, National Institutes of Health grant GM078484, USAMRMC grant W81XWH-10-2-0058, and the Mathers Foundation. Computation used the Extreme Science and Engineering Discovery Environment (XSEDE), which is supported by NSF grant OCI-1053575, and a cluster computer awarded by Silicon Mechanics.
Lobo D, Levin M (2015) Inferring Regulatory Networks from Experimental Morphological Phenotypes: A Computational Method Reverse-Engineers Planarian Regeneration. PLOS Comput Biol, 11(6): e1004295. doi:10.1371/journal.pcbi.1004295.
Tufts University, located on three Massachusetts campuses in Boston, Medford/Somerville and Grafton, and in Talloires, France, is recognized among the premier research universities in the United States. Tufts enjoy a global reputation for academic excellence and for the preparation of students as leaders in a wide range of professions. A growing number of innovative teaching and research initiatives span all Tufts campuses, and collaboration among the faculty and students in the undergraduate, graduate and professional programs across the university's schools is widely encouraged.
The content of this release is solely the responsibility of the authors and does not necessarily represent the official views of the funders.
Return to top of page