In the last years, multiple efforts have been made to accurately predict neoantigens derived from somatic mutations in cancer patients, either to develop personalized therapeutic vaccines or to study immune responses after cancer immunotherapy. In this context, the increasing accessibility of paired whole-exome sequencing (WES) of tumor biopsies and matched normal tissue as well as RNA sequencing (RNA-Seq) has provided a basis for the development of bioinformatics tools that predict and prioritize neoantigen candidates. Most pipelines rely on the binding prediction of candidate peptides to the patient’s major histocompatibility complex (MHC), but these methods return a high number of false positives since they lack information related to other features that influence T cell responses to neoantigens. This review explores available computational methods that incorporate information on T cell preferences to predict their activation after encountering a peptide-MHC complex. Specifically, methods that predict i) biological features that may increase the availability of a neopeptide to be exposed to the immune system, ii) metrics of self-similarity representing the chances of a neoantigen to break immune tolerance, iii) pathogen immunogenicity, and iv) tumor immunogenicity. Also, this review describes the characteristics of these tools and addresses their performance in the context of a novel benchmark dataset of experimentally validated neoantigens from patients treated with a melanoma vaccine (VACCIMEL) in a phase II clinical study. The overall results of the evaluation indicate that current tools have a limited ability to predict the activation of a cytotoxic response against neoantigens. Based on this result, the limitations that make this problem an unsolved challenge in immunoinformatics are discussed.