You should definitely fit your model to the mean data. There is no question about that. However, you have to remember that pooling data runs the risk of masking individual behavior but in your case it does not really matter, since there are no alternative. Your model parameters are therefore more a general guide.

When pooling data, the variability does not say anything about the variability between animals (individual variation) but rather tells you something about the variability between your predicted and your observed data. That is how good your model fits to the average data points.

You might consider checking your model parameters by conducting the experiment for a different dose. However, other (dose-dependent) aspect might chance your model parameters in this case. Therefore, different parameters at different dosages might not be due to an erroneous model. It is therefore important to take animal ethics into consideration before doing such an experiment, since the experiment only gives little additional information.