It struck me a few days ago as I was fiddling with this toy that proteins (and in fact other physical structures) could be thought of as mechanical analogues of neural networks. (It turns out some fascinating intelligent structures can be built with this toy – with some very complex folding behaviors...)
Where in the brain information is propagated via neurotransmitters, in proteins, information is propagated as molecular attractions. Neuron state is the position (orientation, etc) of each amino acid in space.
Proteins could be studied as dynamical networks, and – protein computations could be understood with the same tools neuroscientists use to understand neuronal network dynamics. Furthermore, taking ideas from machine learning, one could effectively train custom proteins to perform bio mechanical functions.
I'm not sure if these ideas are new. What has science said about them?
Dear Akiva
Systems biology as a field is looking into topics like these. You can have a look into terms like protein-protein interaction networks, gene regulatory networks and metabolic dynamics. Kitano is one of the champions in the field:
http://www.nature.com/nature/journal/v420/n6912/abs/nature01254.html
you can also have a look about the relation between systems biology and computational neuroscience:
http://journals.plos.org/ploscompbiol/article?id=10.1371/journal.pcbi.1000078
As mentioned above, computational systems biology have addressed those fundamental questions. Post-genomic view is expanding the protein's role into an element in a network of protein–protein interactions (Jeong et al.,). Systematic mapping of protein–protein interactions (i.e. interactome mapping) was initiated in model organisms, starting with defined biological processes and then expanding to the scale of the proteome (Rual et al.,; Solé et al.,).
These works represent an important step towards a systematic and comprehensive human interactome project.
Article Towards a Proteome-scale Map of the Human Protein-Protein In...
Article Lethality and Centrality in Protein Networks
Article A model of large-scale proteome evolution
These papers are great – thank you so much for taking the time to share them with me. However, I'm looking specifically for whether proteins have been looked at as mechanical neural networks. Do you have anything on that?
Many problems in nature can be deduced to optimization problems. Here you are trying to connect 2 optimization problems:
1. Machine Learning: Minimizing a cost function.
2. Protein Folding: Minimizing free energy.
In theory, you can exchange tools and methods between both fields, but is it practical? I don't know.
Any way, do we have anything as mechanical neural network?
Dr Pimiskern: thank you so much for that paper. It addresses exactly what I was interested in. Separately, for those who come across this in the future I came across this paper by Jurgen Schmidhuber and others which looks at computations with nanomaterials.
I found it interesting to compare the protein logic gate solutions in the protein with the solutions to nanocomputing. I'm not sure if this is by coincidence, but I see similarities:
(purely red solutions are from de Ronde, rainbow solutions from Schmidhuber's paper)
http://doc.utwente.nl/100187/1/future_computing_2016_1_30_30037.pdf
Dear Akiva, this is the very last published attempt to link protein phase spaces and brain activity:
https://www.researchgate.net/publication/301636784_Building_a_minimum_frustration_framework_for_brain_functions_in_long_timescales?ev=prf_pub
Article Building a minimum frustration framework for brain functions...
See also this one:
http://journals.plos.org/plosone/article?id=10.1371/journal.pone.0004154
Hi Akiva,
This seems to fit into the wider idea of using dynamical systems to compute. If you're not already familiar with it, you might be interested in work done in the area of reservoir computing. Here's a paper where they used an actual reservoir of water to do computation, but any dynamical system could be used in principle (and it's more commonplace to use recurrent neural networks):
http://users.sussex.ac.uk/~ctf20/dphil_2005/Publications/bucket.pdf
You might find this paper interesting too: Susan Stepney, "Nonclassical computation: a dynamical systems perspective" -
http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.396.963
Michael
- Badges
- Science topic