• Reduce text

    Reduce text
  • Restore text size

    Restore text size
  • Increase the text

    Increase the text
  • Print


A new protein designed by an Artificial Intelligence

The objective of Artificial Intelligence (AI) is to enable a computer to (better) solve human problems. In Toulouse, an AI using "automated reasoning" algorithms has been able to design an artificial hyper-stable self-assembling protein. This is the result of work involving the MIAT unit, in an informal international collaboration with Belgium and Japan.

The Ika4 protein shown here is the result of the spontaneous assembly of two smaller, identical proteins designed to assemble.
Updated on 12/17/2018
Published on 12/14/2018

Automated reasoning algorithms are targeted at solving extremely complex puzzles involving thousands or millions of interconnected pieces. One of these reasoning AIs has recently been able to prove a theorem that had previously resisted mathematicians for decades. For ToulBar2, the AI that was created at INRA in Toulouse, solving the hardest Sudokus takes only milliseconds. But there are more challenging and more significant puzzles on Earth than Sudokus.

Proteins are the main molecules of life. They govern much of cells' work, in humans, animals, plants, fungi and microbes. They are able to bind to other molecules, and to assemble to build complex structures. They can catalyze chemical reactions at ambient temperature and pressure while being biodegradable. In synthetic biology, teams from Toulouse try to speed up the development of new proteins. While artificial neural networks are the basis for many recent advances in AI, another technology that has been used here is automatic reasoning algorithms.

In collaboration with biochemists from KU Leuven (Belgium) and the RIKEN institute (Japan), Toulbar2 was asked to tackle another puzzle: design a protein by choosing and organizing its atoms so that its copies would self-assemble in water to create a larger symetric protein. Using a preliminary sketch of its structure and an approximate representation of intermolecular forces known as the Talaris14 force field, (from the molecular modeling software Rosetta, Univ. Washington), ToulBar2 faced a puzzle that had more combinations than the number of atoms in the known universe. It was nevertheless able to find an optimal organization of atoms and prove that it was the best possible for Talaris14. No requirement for a super-computer here, a simple desktop computer suffices. The schematics of the designed protein (its amino acid sequence) was then encoded into DNA and inserted in a bacteria (E. coli). The bacteria multiplied and synthesized a large number of copies of the protein. The resulting molecule, called Ika[1], was experimentally verified to have the expected shape and capacity to self-assemble.

Newly designed proteins have a large potential for applications in medicine, green chemistry, biofuels or oil-based products recycling. Having the ability to design new tailored proteins is thus of prime importance for health but also to reduce our environmental footprint.

[1] Because it is made of eight elementary blades, the protein has been called IkA (いか), a Japanese cuttlefish with 8 tentacles.

ToulBar2, the software that solves the most complex puzzles!

ToulBar2 is an AI automatic prover, specialized at solving complex puzzles involving items with discrete values. These problems are hard to solve also for computers (known as NP-complete). An example of a famous NP-complete problem is the Sudoku puzzle. In a Sudoku, every cell holds a number between 1 and 9. All the numbers in a row, columns or 3x3 block must be different. These rules represent properties that must be satisfied by the Sudoku's solution. From these and a Sudoku grid, Toulbar2 will always quickly return an exact solution. The strength of ToulBar2 is that it is not restricted to logical rules. It can also deal with rules involving costs (eg., a rule saying that it's forbidden to have two numbers in adjacent cells differ only by one, with an associated fine for violation of 100 cost units). When such rules exist, ToulBar2 will always produce the least costly solution, and also prove that there is no better one. Thanks to this ability, it can tackle many various real life puzzles. ToulBar2 and its authors learned about and improved protein design thanks to a long lasting collaboration with INSA-LIS.


Noguchi, H., Addy, C., Simoncini, D., Wouters, S., Mylemans, B., Van Meervelt, L., Schiex, T., Zhang, K.Y., Tame, J.R.H. and Voet, A.R.D., 2019. Computational design of symmetrical eight-bladed β-propeller proteins. IUCrJ, 6(1).

Allouche, D., André, I., Barbe, S., Davies, J., De Givry, S., Katsirelos, G., O'Sullivan, B., Prestwich, S., Schiex, T. and Traoré, S., 2014. Computational protein design as an optimization problem. Artificial Intelligence, 212, pp.59-79.

Simoncini, D., Allouche, D., de Givry, S., Delmas, C., Barbe, S. and Schiex, T., 2015. Guaranteed discrete energy optimization on large protein design problems. Journal of Chemical Theory and Computation, 11(12), pp.5980-5989.