This technology utilizes machine learning to accurately predict and prioritize cancer neoantigens by integrating diverse genomic and transcriptomic features beyond traditional peptide-major histocompatibility complex (MHC) binding affinity.
Current approaches to neoantigen prediction rely on peptide-major histocompatibility complex (MHC) binding affinity, overlooking key factors such as antigen processing, transport, and T-cell recognition that are essential for immunogenicity. This leads to limited accuracy in real-world clinical scenarios, leading to ineffective neoantigen selection. As personalized cancer therapies expand, there is an urgent need for more accurate and comprehensive prediction tools to effectively guide target selection and patient stratification. Addressing these shortcomings is critical to improving therapeutic outcomes and streamlining immunotherapy development.
The technology, called ImmuneMirror, uses a machine learning-based computational pipeline to predict and prioritize immunogenic neoantigens by integrating diverse genomic and transcriptomic inputs. Unlike traditional tools that focus mainly on peptide-MHC binding affinity, this tool incorporates critical parameters such as peptide processing, transport efficiency, binding stability, and immune recognition probabilities. ImmuneMirror leverages a balanced random forest model, trained on experimentally validated immunogenic neoantigens. This tool significantly improves prediction accuracy and includes both a standalone pipeline as well as an accessible web server that streamlines neoantigen prediction directly from sequencing data. As such, ImmuneMirror may facilitate the identification and prioritization of therapeutic targets for personalized immunotherapies and enhance patient stratification in clinical oncology trials.
The pipeline has been validated using patient-derived data from gastrointestinal tract cancers, demonstrating improved predictive performance compared to existing methods.