Bayesian multiple instance regression for modeling immunogenic neoantigens.

Area of Science:

• Tumor Immunology
• Computational Biology
• Bioinformatics

Background:

• Understanding tumor neoantigen properties is key to improving cancer immunotherapy efficacy.
• The inefficiency of current immunotherapies highlights the need for better predictive models.
• Neoantigen-specific immune responses remain poorly understood.

Purpose of the Study:

• To develop a novel computational framework for analyzing the relationship between tumor neoantigens and immune responses.
• To identify specific neoantigens that elicit significant immune responses within patient samples.
• To improve the prediction of immunotherapy outcomes by understanding neoantigen immunogenicity.

Main Methods:

• A Bayesian multiple instance regression (BMIR) method was developed.
• The model uses a Gaussian distribution for continuous responses (T cell infiltration) and latent binary variables for neoantigens.
• This approach allows for simultaneous prediction of patient-level responses and identification of key neoantigens.

Main Results:

• BMIR demonstrated superior performance compared to existing optimization-based multiple instance regression methods.
• The method successfully predicts patient-level immune responses and identifies immunologically relevant neoantigens.
• Bayesian statistical inference provides deeper insights into neoantigen-immune interactions.

Conclusions:

• The developed Bayesian multiple instance regression method (BMIR) offers a powerful tool for dissecting neoantigen immunogenicity.
• BMIR advances the understanding of tumor immunology and the development of effective cancer immunotherapies.
• The R package "BayesianMIR" is available for broader application in cancer research.