Using machine learning to predict protein-protein interactions between a zombie ant fungus and its carpenter ant host

Area of Science:

• Mycology
• Neurobiology
• Biochemistry

Background:

• Parasitic fungi employ effector proteins to manipulate host physiology and behavior.
• The zombie ant fungus, Ophiocordyceps camponoti-floridani, alters carpenter ant behavior, causing a distinctive 'summit disease' for transmission.
• Interspecific protein-protein interactions (PPIs) are hypothesized to mediate this host manipulation.

Purpose of the Study:

• To predict and analyze host-parasite PPIs between O. camponoti-floridani and Camponotus floridanus using machine learning.
• To generate mechanistic hypotheses for how the fungus manipulates ant behavior.
• To identify potential fungal effector proteins and their host targets.

Main Methods:

• Utilized D-SCRIPT, a machine learning tool, for high-throughput prediction of PPIs between the parasite and host.
• Analyzed predicted PPIs for functional enrichment in both host and parasite proteins.
• Focused on parasite proteins encoded by genes upregulated during manipulated host behavior.

Main Results:

• Predicted approximately 6000 PPIs involving 2083 host and 129 parasite proteins.
• Identified overrepresented functional annotations in host proteins, including neuromodulatory G-protein coupled receptors and oxidation-reduction processes.
• Found enrichment of parasite proteases and novel small secreted proteins, suggesting potential effector roles.

Conclusions:

• The study provides novel hypotheses regarding the molecular mechanisms of zombie ant behavior manipulation.
• Identified specific host targets (e.g., G-protein coupled receptors) and parasite effectors (e.g., proteases, small secreted proteins).
• Highlights the utility of machine learning for predicting interspecific interactions in host-parasite systems.