Yes, the data I get from my computer is usually quite accurate. We use data on *a lot* of people (one of our studies has over a million people in it) to make our results as accurate as we can get them to be.
In any computer simulation of reality, we make simplifying assumptions to make the calculation faster. As long as we know the extent to which those simplifications may introduce an error into the results, we can say how far we can trust our results. In general, the more simplifications, the greater the error compared to reality. That is why we try to back up our simulated data with real, experimental data, so see how well they match each other. In fact, I was talking to someone who worked for a simulation company the other day, and he told me that the people who write the simulation software will often do experiments as they are writing the program – to ensure that they are writing more realistic behaviour into the program.
Thats a question that in never far from my mind. When I develop a model, before I use it to try and work out things that are unknown, I absolutely must validate it. At the moment, I have a data set of 124 different molecular machines for which I know how they assemble. These are my test cases, and they are very interesting in themselves. One, for example, is an antibody which attaches itself to birch pollen and can trigger hayfever. Another, called NEF, is used by HIV to avoid our immune system whilst a third (called 14-3-3), is made in the brain and tells us when to go to sleep!
Anyway, when I test out a new idea, I apply it to these 124 test cases and see if it works. This is how I know if my models are accurate. At the moment, they are pretty good. I think they are about ready to start putting to use to see how cancer-related molecular machines assemble themselves.