MHC class I antigen processing consists of multiple steps that result in the presentation of MHC bound peptides that can be recognized as T cell epitopes. Many of the pathway steps can be predicted using computational methods, but one is often neglected: mRNA expression of the epitope source proteins. In this study, we improve epitope prediction by taking into account both peptide-MHC binding affinities and expression levels of the peptide’s source protein. Specifically, we utilized biophysical principles and existing MHC binding prediction tools in concert with RNA expression to derive a function that estimates the likelihood of a peptide being presented on a given MHC class I molecule. Our combined model of Antigen eXpression based Epitope Likelihood-Function (AXEL-F) outperformed predictions based only on binding or based only on antigen expression for discriminating eluted ligands from random background peptides as well as in predicting neoantigens that are recognized by T cells. We also showed that in cases where cancer patient-specific RNA-Seq data is not available, cancer-type matched expression data from TCGA can be used to accurately estimate patient-specific gene expression. Using AXEL-F together with TGCA expression data we were able to more accurately predict neoantigens that are recognized by T cells. The method is available in the IEDB Analysis Resource and free to use for the academic community.Significance statementEpitope prediction tools have been used to call epitopes in viruses and other pathogens for almost 30 years, and more recently, to call cancer neoantigens. Several such tools have been developed, however most of them ignore the mRNA expression of the epitope source proteins. In the present study, we have, to our knowledge for the first time, developed a biophysically motivated model to combine peptide-MHC binding and abundance of the peptide’s source protein to improve epitope predictions. Our novel tool AXEL-F is freely available on the IEDB and presents a clear opportunity for predicting and selecting epitopes more efficiently.