In genetics and evolutionary theory, the fitness landscape is often used as a metaphor. Here, instead, we treat it as a physical entity in which genomes move under mutational displacements. This model enables the definition of an energy term for genome variants, computable for coding mutations. Surprisingly, coding variants found in healthy individuals and populations seem to follow elementary laws of thermodynamics. However, in patients with complex diseases, a subset of genes carries variants that do not. This divergence in behavior was seen in Alzheimer’s disease, autism, and cancer, each time pointing to genes known to be involved in disease risk, or likely to, even when using smaller populations. These findings establish a new and formal bridge between physics, genetics, and personalized medicine by way of the fitness landscape.