Slime as Engineer - brainless mold mimics Tokyo subway

slimemold.jpg Physarum polycephalum + oat flakes = Tokyo subway map?
Because they couldn't mathematically determine a "perfect" solution, the researchers decided to task the slime mold with a problem human designers had already tackled. They placed oat flakes (a slime mold favorite) on agar plates in a pattern that mimicked the locations of cities around Tokyo and impregnated the plates with P. polycephalum at the point representing Tokyo itself. They then watched the slime mold grow for 26 hours, creating tendrils that interconnected the food supplies. Different plates exhibited a range of solutions, but the visual similarity to the Tokyo rail system was striking in many of them
[via jetlib express, abstract of full article]


  1. The full article is even more interesting, if you’re able to read it (the benefits of a university net connection). They use light and shade to simulate topography, to encourage the slime to build in low areas rather than high ones, and to avoid crossing water much.

    I really like this as a novel solution to transport planning!

  2. It’s wrong to say that the train network of Tokyo was built to efficiently connect places where people lived.

    Instead, the train network was built to sell real estate, so people live where the train stations are, and the network was built where people *didn’t* yet live.

    1. No need for LOLing. This sort of approach is often explored for NP-complete problems where near-optimal solutions are sufficient. Normally we simulate the “slime mold”.

      See for a general approach where the optimal solution is encoded as a lowest-energy state in a solid, and we basically use temperature variation to “grow a crystal” which solves the problem. :-)

  3. awesome. does this count as an empirical rather than a theoretical solution?

    and if so, does asking another human what they think about a problem count as an empirical rather than theoretical solution?

    how about a human with which you do not share a language, but whose solution you are able to see the logic of?

    pretty cool.

  4. So I’m guessing that using a sentient black pudding as one’s janitor is the natural extension of this?

    1. And have it kill the security chief and torture the first officer?! Let’s not get ahead of ourselves.

  5. What’s rather amusing to me is the this is a classic “traveling salesman” problem, mathematically fairly intractable. Such problems are one of the things that quantum computers will be really, really good at solving, it’s been suggested. However slime mold, oatmeal, and a petri dish with substrate seems to be doing a pretty good job here!

    Now all we need it to get it to solve factorization problems!

  6. Looks like this is a biological example of constructal theory, a very interesting concept that comes out of thermodynamics and explains heat transfer, tree structures, and traffic patterns, among many other things.

  7. Ha! we talked about this at journal club yesterday- worth mentioning that it is not a subway map but a national railway map. the study has its issues, but the authors don’t make outlandish claims, and we all thought it was a fun story! The problem as an MST isn’t difficult to solve, but the biological approach is interesting. I wish MBTA would consult slime mold for a change…

  8. @#2 is correct. This looks like coincidence deriving from the fact that the researchers pre-placed the locations of the stations as if they made some kind of sense in themselves. Having done this, it’s hardly surprising that the links that evolve between them look at bit like the Tokyo network. But the thing is that the locations of the stations were not givens. The Tokyo railway system was not constructed by some rational plan but though several often conflicting mechanisms. Firstly, a lot of terminals and the lines into them were built by private companies in order to generate business for their department stores and other facilities. The same companies built housing in order to create a captive market. Then there was the influence of modernisation, of earthquakes and destruction in war, of land prices (which were themselves influenced by railway line and station location, as well as global capital markets) and so on, and so on. Sometimes we should step back and realise out human pattern recognition systems cause us to over-interpret superficial resemblances.

  9. I just finished reading ‘Prey’ by Michael Crichton, and he mentions this more than once in the book.

  10. Nitpick: not actually a brainless mold (Fungi), despite the name, but actually a brainless amoeba (Amoebozoa).

  11. I think a good deal, if not most, of what’s being shown in the simulation would not be subway. The big yellow ball in the center would in fact encompass most of the Tokyo subway system. The article says “rail system” and does not make a reference to subways.

  12. I heard of something similar using humans and a newly-built university campus: the designers didn’t put the paths between buildings in until the students had been on campus for a few months during which time they of course walked between the buildings – the designers then paved over the worn out bits of grass etc that denoted where people wanted to walk. I don’t know if the paths were always the shortest routes or whether topography had an effect but it seemed like a sensible way to go about things when I heard about it.

    1. @Patrick, as far as I know humans do not necessarily take the shortest route due to the fact how they navigate. Apparently just putting a fountain or something in the the middle of a large open space will alter the routes, as humans will use that one for navigation, i.e. work first towards the nearer, intermediate goal and switching goals later.

  13. What’s really fun is that Physarum’s network transports nutrients both ways in those tubes but never at the same time, like a pumping station. I’m sure if they used fluorescent antibodies on some proteins, they could study natural traffic flow.

    Here’s a closeup under a dissecting microscope. I was a freshman at the time so sorry about the quality:

  14. My grandfather told me that when he was charged to build a small road from the existing road (itself a one-lane horse path) up the hill to a wealthy person’s new home, he set a horse with a wagon free ( I don’t know what the inducement to the horse to walk up the hill was – food, promise of a 40 hour work week, date with another horse, tickets to the Boston Braves) but he kept track of where the horse walked and chose that winding and interesting route for the road.

  15. They should get some slime molds to actually manage the Chicago Transit Authority and it would work better and be less corrupt than with the clowns who run it now.

  16. The problem I *REALLY* want to see is a complex networking one.

    How do you setup a NOC with, say, 200 racks, two floors, each rack, say, 32 machines, and each row being like the APC heated center design, 10 racks on each side, with interconnection between all of them and an outgoing point of presence, with the main concern being high availability?

    (also how to distribute the data, but at least the slime might be of use for the network side of the problem, it’s all one MASSIVE cluster, running a single OS instance that is fault tolerant.)

Comments are closed.