Genome sequencing has significantly improved diagnosis and treatment of genetic diseases and cancers; however, it produces many variants that cannot be easily interpreted. Automated bioinformatics tools can help predict functional consequences of variants of uncertain significance, but their accuracy is low. Here, by tracing precise evolutionary history of each amino acid position in MEK1 kinase (mutations in which cause neurodegenerative diseases and cancer), we can establish whether variants seen in humans are evolutionarily tolerant. Using published data and experiments, we show that evolutionarily tolerable variants in MEK1 are benign, whereas intolerable substitutions are damaging. Our approach will help in diagnostics of MEK1-associated diseases, it is generalizable to many other disease-associated genes, and it can help improve automated predictors.