P051 A machine learning model that predicts binding between KIR3DL1 and HLA class I allotypes

AIM KIR3DL1 is a polymorphic inhibitory NK receptor that recognizes HLA class I allotypes that contain the Bw4 motif. Structural analyses have shown that in addition to residues 77–83 that span the Bw4 motif, polymorphism at other sites throughout the HLA molecule can influence the interaction with KIR3DL1. Given the extensive polymorphism of both the KIR3DL1 and HLA class I we sought to train and evaluate a model for predicting the binding between any combination of KIR3DL1 and HLA class I allotypes. METHODS KIR3DL1 tetramers were screened for reactivity against a panel of HLA-I molecules which revealed different patterns of specificity for each KIR3DL1 allotype. This data was used to train several machine learning models (Support Vector Machine, Multi-Label Vector Optimization, Linear Regression, Neural Network) to learn the association between the amino acid sequence of the HLA class I allotype and the normalized MFI of the tetramer binding experiments. Separate models were trained for each of 6 KIR3DL1 allotypes. RESULTS The performance of the predictor was evaluated by random cross-fold (80/20) validation and computing the area under the curve (AUC) of the receiver-operator characteristic (ROC). The MLVO model performed best with AUC scores ranging from 0.795 to 0.846 for the 6 KIR3DL1 allotype models. CONCLUSIONS This binding predictor can be applied to clinical datasets to more specifically investigate the role of HLA-KIR interaction and improve clinical decision making.