Toward equitable major histocompatibility complex binding predictions.

Area of Science:

• Immunoinformatics
• Computational Biology
• Genomics

Background:

• Major histocompatibility complex (MHC) binding prediction is crucial for personalized cancer immunotherapies and vaccines.
• Equitable application requires methods that perform well across diverse human populations and MHC alleles.
• Significant disparities exist in the amount of binding data available for different MHC alleles across racial and ethnic groups.

Purpose of the Study:

• To assess the impact of data imbalance on MHC binding prediction across diverse populations.
• To develop an advanced MHC binding prediction model that accounts for and mitigates data disparities.
• To establish a foundation for equitable MHC binding models in personalized medicine.

Main Methods:

• Developed a machine learning framework to evaluate data imbalance effects on MHC allele binding predictions.
• Applied the framework to create a state-of-the-art MHC binding prediction model with per-allele performance estimates.
• Devised an algorithmic strategy for targeted data collection to address remaining inequities.

Main Results:

• Identified alarming disparities in MHC binding data availability across racial and ethnic groups.
• The developed MHC binding model significantly mitigates performance disparities linked to data imbalance.
• The model provides per-allele performance estimates, enhancing transparency and reliability.

Conclusions:

• The study highlights critical data equity issues in MHC binding prediction for immunotherapies.
• The novel framework and model advance the development of fairer and more accurate predictive tools.
• This work is foundational for creating equitable MHC binding models essential for personalized cancer treatments.