Cancer Immunotherapies Ignited by a Thorough Machine Learning-Based Selection of Neoantigens

Area of Science:

• Oncology
• Immunology
• Bioinformatics

Background:

• Neoantigens from somatic DNA alterations are key targets for cancer immunotherapies.
• Predicting neoantigen immunogenicity is challenging due to variable mutation effects.
• Efficient tools are needed to identify immunogenic neoepitopes.

Purpose of the Study:

• To develop and present a comprehensive computational pipeline for identifying neoepitopes.
• To predict the immunogenicity of neoepitopes based on genomic sequencing data.
• To enable both selection from patient data and generation of novel neoantigen candidates.

Main Methods:

• A multi-step pipeline integrating data pre-processing and three machine learning predictive models.
• Analysis of genomic alterations, human leukocyte antigen (HLA) typing, and T-cell receptor (TCR) repertoire.
• Machine learning models predict neoantigen classification, binding affinity to major histocompatibility complex class I (MHC-I), and immune response likelihood.

Main Results:

• The pipeline successfully identifies neoepitopes with high potential for immunogenicity.
• It can be applied to select neoantigens from patient sequencing data.
• Novel neoantigen candidates can be generated using Monte Carlo and Reinforcement Learning techniques.

Conclusions:

• The presented pipeline offers a robust solution for neoantigen identification and prediction.
• This approach can significantly advance the development of personalized cancer immunotherapies.
• The generative capabilities of the pipeline open new avenues for neoantigen discovery.