Table 1.
Demographic/Ecological Parameters.
Table 2.
Gene Drive Parameters.
Fig 1.
Gaussian Process model schematic.
The training set started as 1000 randomly sampled points, and the models were considered complete after the training set had been iteratively grown to 10000 points.
Fig 2.
Population dynamics without gene drive release.
Left panel: Population size fluctuations over time are shown after the population is initialized at the approximated capacity of 4000. The average population size was 4257 individuals. Center panel: The average frequency of each age, as determined by tabulating the ages of each individual between time step 100 and 500 of the simulation. Right panel: Ripley’s L at length scales from 0 to 200 meters. This plot shows the difference between Ripley’s L statistic in the simulation and the expected value of a random distribution. Negative values indicate that a population is more dispersed than a randomly distributed population. Density dependent competition takes place at a maximum distance of 75 meters and is more intense between close-together individuals, resulting in a slightly dispersed population.
Fig 3.
Drive efficiency required to eliminate the population using the default model.
Drive efficiency was varied in 1% increments, leaving other parameters at default values, with 100 simulations conducted at each point. A “successful run” is defined as a simulation in which the population is completely eliminated within 500 time steps.
Fig 4.
Time from release until suppression for the drive targeting female fertility.
Number of time steps until population eradication for the homing drive with a female fertility target using default parameters and a drive efficiency of 95% (blue) and 85% (orange). One thousand simulations were performed for each drive.
Fig 5.
Gene drive spread over time for the drive targeting female fertility.
A series of snapshots showing the progress of a single simulation of the homing drive with a female fertility target using default parameters and a drive efficiency of 95%. Black dots are wild type, green dots are drive/wild-type heterozygotes, and blue dots are drive homozygotes. The upper left panel depicts the time step at which the drive was introduced to the population. At first, the drive rapidly spreads through the population (upper right). After the drive has reached a sufficient frequency, the population dwindles in size (lower left) until almost all possible pairings only produce sterile offspring (lower right), leading to population elimination shortly thereafter.
Table 3.
Gaussian Process Model Accuracy.
Fig 6.
Model quality, female fertility homing drive without resistance.
The model was evaluated against the 10000-point Latin Hypercube sampled test set. This drive has a larger area of success than all of the other drives, and thus, the GP model performs relatively well, even before adaptive sampling, compared to the other models. The first few adaptively sampled data sets substantially improved the performance of the model, after which improvements were smaller. See S1 Fig for similar plots for the other models.
Fig 7.
Model comparison, female fertility homing drive without resistance.
For this comparison, drive fitness and efficiency were varied while other parameters were kept at default values. A 1000 by 1000 array of queries was made to both GP models; color indicates model predictions of drive success or failure, as well as confidence. Black and white square dots show results from simulations using the underlying model. Each square dot denotes the result of twenty simulations performed at each point. This overlay serves as a visual validation of the GP framework; the actual model behavior in this parameter range (i.e., how drive success depends on model parameters) is discussed in more detail in the “Selected Model Outcomes” section below. See S2–S5 Figs for similar plots for the other models.
Fig 8.
“First order effects” describe the effects of varying a single parameter. “Total effects” include the first order effects of the parameter, as well as potential synergistic interactions between that parameter and one or more other parameters. Analyses of drive models without resistance sampled 24 million points from the parameter space. Analyses on drive models with resistance sampled 28 million points. These analyses are from the models trained on the suppression rate due to their somewhat better accuracy. Analyses of the composite model were similar.
Fig 9.
Sobol sensitivity analysis with fixed fitness and efficiency.
For the female fertility homing drive, fitness and efficiency were both fixed at 90%. For the viability homing drive, fitness and efficiency were both fixed at 97%. For the Y-shredder, fitness and efficiency were both fixed at 99%. The models without resistance were used. “First order effects” describe the effects of varying a single parameter. “Total effects” include the first order effects of the parameter, as well as potential synergistic interactions between that parameter and one or more other parameters.
Fig 10.
Comparison of the three drives with varying fitness and efficiency.
Other parameters were fixed at default values. For each panel, a 1000 by 1000 array of queries was made to the GP model for that drive.
Fig 11.
Comparison of the three drives with varying litter size and migrant frequency.
For the female fertility homing drive, fitness and efficiency were both fixed at 90%. For the viability homing drive, fitness and efficiency were both fixed at 97%. For the Y-shredder, fitness and efficiency were both fixed at 99%. The models without resistance were used.
Fig 12.
Female fertility homing drive with varying fitness, efficiency, and survival rate.
Other parameters were fixed at default values.
Fig 13.
Y-shredder with varying interaction distance and average dispersal.
Drive fitness and efficiency were both fixed at 99%. Other parameters were fixed at default values.
Fig 14.
Comparison of the two homing drives with varying drive fitness, efficiency, and resistance parameters.
Survival rate was set to 0.8, and other parameters were fixed at default values.