Newswise – Marine Corn Snails are hosts of the dangerous family of neurotoxins. Little is known about how these toxins interact with the human body, which has become an area of interest in drug research and the national security space. For the first time, the team at Los Alamos National Laboratory has successfully trained a machine learning model to predict how alphaconotoxins bind to specific human receptor subtypes.
“It is estimated that only 2% of them have been sequenced due to the diversity and complexity of natural conotoxins,” says Gunana Gunanakaran, a theoretical biologist at Los Alamos. “Conotoxins do not have antidotes, but we have the ability to develop tools to understand and respond to these threats by using machine learning to predict conotoxin binding.”
The lethal secretion issued by any of the more than 800 corn snail species represents an assemblage of over 1 million natural conotoxins. The researchers focused their machine learning work on alpha conotoxins, a common and deadly family of conotoxins.
The machine learning model implemented by the team overcomes the limited data challenges for incorporating alpha conotoxin amino acid sequences, secondary structural tendencies, and electrostatic properties, and successfully predicts human receptors containing receptor subtypes – toxins will be targeted. This is the first known machine learning model to achieve that. In addition to Gnanakaran, the team that included lead author Hung Nguyen Do and biochemist Jessica Kubicek-Sutherland presented their results at ACS Chemistry Neuroscience.
Predicting subtype-specific targets
Conus Geography, also known as Geography, claims the distinction between the most lethal corn snails. The 65% lethality rate from a single prick is driven primarily by the activity of alphaconotoxins, which binds to human nicotinic acetylcholine receptors, and is integrated into human muscular, nervous and tissue functions. Binding to these receptors inhibits function that often results in lethal consequences.
The challenge in predicting conotoxin binding is the lack of data on receptor binding across different conotoxin families. To understand which receptor subtypes are bound by alphaconotoxins, the researchers developed and deployed two different neural network architectures to train semi-surveillance machine learning models.
The team determined that the most effective semi-spived machine learning classifier, the most effective way to predict target receptor subtypes, came from training neural network architectures with a combination of density and convolutional layers on amino acid sequences, secondary structure, and electrostatic properties.
Next step: Use predictions in your experiments
The team's next step is to take these predictions into the experimental stage of the Los Alamos chemistry lab and create an actual interface that can mimic conotoxin interactions and binding. Such an interface could be made to act as an antitoxin. This is a useful tool considering the possibility that a bad actor develops synthetic neurotoxin.
Neurotoxin-like agents can potentially develop in synthetic peptides (short chains of amino acids) and exhibit similar behaviors and effects as seen with naturally occurring conotoxins. Unfortunately, the development of synthetic peptides that behave in the same way as conotoxins requires much more understanding than antitoxin development. But with the model in hand, researchers can apply insights to the problem.
“No one has asked these questions about binding to target receptor subtypes, but it's gone before,” Kubicek-Sutherland said. “The experimental stage of this allows us to collect models developed through artificial intelligence tools and see if they are really working. That work is an important step in developing effective antitoxins.”
paper: “Predicting the specificity of α-conotoxins for human nicotinic acetylcholine receptor subtypes by semi-surveillance machine learning.” ACS Chemistry Neuroscience. doi:10.1021/acschemneuro.4C00760
Funds: This work was supported by the US Department of Defense Threat Reduction Agency.
LA-UR-25-26652
