EpLogic

Scripts for Anti-HLA antibody target prediction via machine learning.

Description

These datasets comprise mismatched eplets from HLA alleles tested in single antigen panels of patients waiting for a solid organ transplantation. The experiments described here have the aim of classifying each eplet/panel/patient as reactive or non-reactive. Tip: reactive eplets usually have greater MFI values.

Experiments

• Simple train/test validation. 89% accuracy and 88% AUC-ROC.
• Cross-validation. 88% accuracy and 92% AUC-ROC.
• Cross-validation & feature selection. 89% accuracy and 94% AUC-ROC.
• Cross-validation & feature selection & hyperparameter tuning. 91% accuracy and 94% AUC-ROC.

Support

You can contact me at mariomarroquim@gmail.com.