Because these two toxicities work through completely different mechanisms, the researchers approached them separately.
block neurotoxins
Neurotoxic three-finger proteins are a subgroup of a larger family of proteins that specialize in binding to and blocking receptors for the major neurotransmitter acetylcholine. Their three-dimensional structure, which is key to their ability to bind to these receptors, is based on three strings of amino acids that nestle together within the protein (for those who have taken a sufficiently advanced biology course, these are opposites). parallel) beta sheet). So, to thwart these toxins, researchers targeted these strings.
They relied on an AI package called RFdiffusion (RF indicates its relationship to Rosetta Fold protein folding software). RF diffusion can be directed to the design of protein structures that complement specific chemicals. In this case, a new chain was identified that could line up along the end of the three-finger toxin chain. Once they were identified, another AI package called ProteinMPNN was used to determine the amino acid sequences of the full-length proteins that form the newly identified chains.
However, the use of AI tools is not yet over. The combinations of the three-finger toxin and the newly designed set of proteins were then input into DeepMind’s AlfaFold2 and Rosetta protein structure software to estimate the strength of the interaction between them.
It was only at this point that the researchers began creating the actual proteins, focusing on candidates that the software suggested would interact best with the three-finger toxin. Forty-four of the computer-designed proteins were tested for their ability to interact with the three-finger toxin, and the single protein that showed the strongest interaction was used for further studies.
At this point, we returned to AI and used RFDiffusion to suggest variants of this protein that might bind more effectively. In fact, about 15% of the suggestions interacted more strongly with the toxin. The researchers then created both the toxin and the most potent inhibitor in bacteria and obtained the structure of their interaction. This confirmed that the software’s predictions were highly accurate.
