Rapidly increasing pyrethroid insecticide resistance and changes in vector biting and resting behavior pose serious challenges in malaria control. Mosquito repellents, especially spatial repellents, have received much attention from industry. We attempted to simulate interactions between mosquitoes and repellents using a machine learning method, the Self-Propelled Particle (SPP) model, which we modified to include attractiveness/repellency effects. We simulated a random walk scenario and scenarios with insecticide susceptible/resistant mosquitoes against repellent alone and against repellent plus attractant (to mimic a human host). Simulation results indicated that without attractant/repellent, mosquitoes would fly anywhere in the cage at random. With attractant, all mosquitoes were attracted to the source of the odor by the end. With repellent, all insecticide-susceptible mosquitoes eventually moved to the corner of the cage farthest from the repellent release point, whereas, a high proportion of highly resistant mosquitoes might reach the attractant release point (the human) earlier in the simulation. At fixed concentration, a high proportion of mosquitoes could be able to reach the host when the relative repellency efficacy (compare to attractant efficacy) was <1, whereas, no mosquitoes reached the host when the relative repellency efficacy was > 1. This result implies that repellent may not be sufficient against highly physiologically insecticide resistant mosquitoes, since very high concentrations of repellent are neither practically feasible nor cost-effective.

Rapidly increasing pyrethroid insecticide resistance and changes in vector biting and resting behavior pose serious challenges in malaria control. Mosquito repellents, especially spatial repellents, have received much attention from industry. Many of these repellents contain the same or similar chemicals as those used in insecticides. Does resistance to insecticides affect the efficacy of spatial repellents? We attempted to simulate interactions between mosquitoes and repellents using a machine learning method, the self-propelled particle (SPP) model, which we modified to include attractiveness/repellency effects. We simulated a random walk scenario and scenarios with insecticide susceptible/resistant mosquitoes against repellent alone and against repellent plus attractant (to mimic a human host). We assumed attractant odors and repellent chemicals diffused randomly and omnidirectionally, and that mosquitoes were confined in a cubic cage. We modified the velocity and direction components of SPP using attraction/repulsion rates and concentrations. Simulation results indicated that without attractant/repellent, mosquitoes would fly anywhere in the cage at random. With attractant, mosquitoes might or might not fly toward the source (i.e., the human), depending on the simulation time (and thus the attractant concentration distribution). Eventually, however, all mosquitoes were attracted to the source of the odor. With repellent, results depended on the amount of chemical used and the level of mosquito insecticide resistance. All insecticide-susceptible mosquitoes eventually moved to the corner of the cage farthest from the repellent release point. Surprisingly, a high proportion of highly resistant mosquitoes might reach the attractant release point (the human) earlier in the simulation when repellent was present compare to no repellent was present. At fixed concentration, a high proportion of mosquitoes could be able to reach the host when the relative repellency efficacy (compare to attractant efficacy) was <1, whereas, no mosquitoes reached the host when the relative repellency efficacy was > 1. This result implies that repellent may not be sufficient against highly physiologically insecticide resistant mosquitoes, since very high concentrations of repellent are neither practically feasible nor cost-effective.