Machine learning enhanced immunologic risk assessments for solid organ transplantation.

Area of Science:

• Immunogenetics
• Transplant immunology
• Computational biology

Background:

• Solid-organ transplantation requires precise recipient-donor compatibility matching.
• Human Leukocyte Antigen (HLA) allele and antibody profiling are critical for predicting transplant outcomes.
• Current methods for crossmatch compatibility prediction are complex and can be improved.

Purpose of the Study:

• To enhance the prediction of solid-organ recipient and donor crossmatch compatibility using machine learning (ML).
• To develop an ML model that determines donor-specific antibodies from a recipient's full HLA antibody profile without laboratory interpretation.
• To improve personalized histocompatibility risk assessments in transplantation.

Main Methods:

• Developed an HLA allele imputation system to convert HLA antigens to alleles.
• Combined imputed and known HLA alleles with recipient HLA antibody profiles.
• Applied various ML models to predict crossmatch reactivity based on donor-specific antibodies.

Main Results:

• Achieved a high prediction accuracy with an ROC-AUC of 0.975.
• Demonstrated the ML models' ability to predict crossmatch reactivity effectively.
• Provided insights into the significance of specific HLA antibodies in transplant matching.

Conclusions:

• ML models can significantly enhance the prediction of solid-organ transplant crossmatch compatibility.
• The developed approach offers a powerful tool for personalized histocompatibility risk assessment.
• This study advances the understanding of HLA antibody profiles in predicting transplant success.