One of the more exciting and important developments in recent years is personalized cancer vaccines. We sequence the tumor of each patient. We synthesize the fragments of the proteins that are mutated, and administer them to patients in a shot. This stimulates the immune system to attack the cancer by teaching what is the cancer.
This was working for patients, in many cases. We, however, realized a problem. The process of inferring the mutated protein, lays mutations of the patient on a default sequence. The reference genome was being used as a the default sequence from which to compute mutated protein fragments, rather than the patients own genome. In a surprisingly high number of patients, they were getting the wrong vaccine. This paper fixes this problem.
Happy to answer any questions here about this work.