Talk: Alife 2020 keynote Michael Levin - Robot Cancer

Emergence, Biological life, ALife 2020

How do organisms store information and are able to pass it down through very profound structural changes (from a caterpillar to a butterfly, when cutting a flatworm in multiple pieces, etc.)?

Embryogenesis is a reliable self-assembly. It relies on stem cell differentiation but that’s not enough: some tumor (Teratoma) are differentiated but don’t have the right 3D spatial organisation.

Where is the large-scale pattern specified?

Everything has to be in the right shape with the right organization. The typical answer is DNA. However DNA is really just protein structure specification and not a blueprint for anatomy.

How do cell groups know what to make and where to stop?

Can we control the process to repair and edit, and can we build arbitrary things with that process?

Current anatomy paradigm

Gene regulatory networks (turn genes on and off). Some genes become effector proteins that interact via physics and all of this happens in parallel and an organism emerges.

However it is very difficult to figure out the inverse transformation: figure out the molecules to tweak to get a particular organism.

The process is reliable but very adaptive (not hardwired). It can regulate itself after drastic perturbation such as splitting an embryo in two gives twins. We can also aggregate embryo into a single organism.

Many organisms can also do that at the adult stage (e.g. axolotl regenerate limbs, eyes, jaws, etc.).

One can graft a tail to the leg of a salamander and it can remodel itself into an arm.

Planarians are really great at that. When cutting in the middle, cells from that middle position will make very different choices and regrow the right part. These organisms are immortal.

Regeneration is not just for these organisms: human liver, deer regenerate bones and skin every year, human children can regenerate fingertips.

The body is very plastic, it can go towards specific anatomical states through self-assembly. It is not a local behavior, there is a collective decision in cell swarms.

Individual cells can be very competent, and with cooperation they became even more competent.

Tadpoles transform into correct frogs even when all the organs are misplaced. The anatomy program is very robust and able to react to unexpected situations and correct errors.

It is possible there might be a homeostatic set-point (or pattern memory). Then if we understand it we might be able to edit it and control the blueprint without touching the cells themselves. The endgame would be an anatomical compiler, in which we could design the overall layout of the target organism and let it develop according to the blueprint.

For now we are good at understanding gene regulatory networks, but ultimately we would wand to understand how this can get us to organs.

Interesting parallel with the computer science revolution, we historically focused on hardware but later turned to the actual algorithms and software going higher and higher up in the abstractions. If biology is a programmable hardware, we should be able to program it.

Non-neural bio-electricity

In the brain, genes specify electrical circuits, ion channels and synapses (this is the hardware). The software is all the activations within these neurons which lead to pattern recognition, memory, etc.

But actually, all cells have ion channels and a form and electrical synapses therefore there is a developmental “software” to be discovered too.

The speaker’s group have developed tools to control the bio-electric network in vivo by using molecular actions on the synapses and ion channels. This enables coherent large-scale changes to the anatomy. The authors can build an eye anywhere in an organism. . Different developmental states in flat worms correspond to different attractor state of the electrical circuits which can be controlled without touching the genome. Because of these different attractors, one can direct development into an attractor evolution doesn’t use.

This can be used for reconstruction. Example of tadpoles with mutated gene which prevent the from growing a brain. By controlling bioelectric signals one can regrow the brain despite the mutation.

Pattern memory

By editing the bioelectric patterns one can make a planarian grow as two-headed. Therefore, the bioelectric pattern isn’t an indicator of what’s happening now in the anatomy, it encodes the pattern that will guide anatomy if it is cut.

What about when we don’t actively perturb the bioelectric pattern? Shouldn’t the two-headed worm go back to normal if cut again? Actually, because the other attractor exists in the electric network, the work can keep regrowing two heads in perpetuity, or steered back to its normal form. This has all the properties of memory.

In this space of networks, one can see the memory of particular anatomical state as attractors, or basins of attractions of the neural net dynamics.

Slide from Michael Levin's talk

Figure 1: Slide from Michael Levin’s talk

DNA encodes an excitable medium with symmetry-breaking dynamics (the system does something if nobody disturbs it) which is re-writable because software is modular and controls anatomy.

Multi-scale organization and ALife

Cognitive agents are made of parts. Bodies are swarms of cells, and morphogenesis is a great framework for studying it.

Molecular networks give cells which give organs which give organisms and finally societies. Autonomy happens at those multiple scales. And because of this organisms are evolvable.

Links to this note

Last changed | authored by


← Back to Notes