This is not a simple or solved problem. In fact, it's one of the greatest challenges facing the field today. Perhaps the most robust way to find drivers is leveraging large cohorts of samples and using recurrence as an indicator of selection (relevant methods: the SMG test in MuSiC or MutSig). Another is to use functional annotation to infer driver status. If a mutation activates a kinase, or alters a gene in a tumor-suppressor pathway, it's possibly important.

That said, there are undoubtedly a large number of rare mutations that contribute to oncogenesis that have not been discovered or understood. As I said, it's a very difficult problem.

Quick, before anyone answers this question make sure to publish your answer in a high-profile journal for fame and fortune!

I think you'll be very interested in this research group methods and tools

They've recently published a series of very relevant papers related to your question.

Hi, to follow up a bit on Chris' answer, I would add a semantic twist to your question:

• Short answer : Use databases indicating whether a mutation can be a driver, e.g. COSMIC etc.
• intermediate answer : You can't answer your question just by observation
• Long answer : You can determine whether you have sufficient evidence to consider that a mutation is likely a driver mutation (e.g. if it's significantly often mutated in cancers), but that'll always only be evidence for follow-up experiments, where you'll e.g. induce the specific mutation and e.g. observe proliferation advantages. Apart from that, you can only recover indications that a gene / mutation might be driver.

The review by Bert Vogelstein and the Hallmarks of cancer papers can help you clarify this as well as give additional definitions and indications.