Molecular representation learning (MRL) is a crucial link between machine learning and chemistry. It plays a vital role in predicting molecular properties

Objective: This study aimed to investigate whether hippocampal texture asymmetry features extracted from PET could effectively differentiate anti-LGI1 encephalitis from anti-GABABR encephalitis based on a hippocampal laterality radiomics (HLR) model. Methods: A total of 82 patients (57 anti-LGI1, 25 anti-GABABR) were retrospectively enrolled. Radiomic texture features were extracted from the left, right and bilateral hippocampus. Asymmetry features were calculated based on hippocampal texture features, which were used to construct HLR model. In addition, a left hippocampal radiomics (LHR) model, a right hippocampal radiomics (RHR) model, and a bilateral hippocampal radiomics (BHR) model were constructed based on texture features of left, right and bilateral hippocampus, respectively. The leave-one-out cross-validation was utilized for hyperparameter tuning. Model performance was evaluated using receiver operating characteristic (ROC) curve analysis, decision curve analysis (DCA), integrated discrimination improvement (IDI), and net reclassification improvement (NRI), and SHAP feature analysis. Results: The HLR model achieved the best predictive performance, with an AUC of 0.898, accuracy of 84.21%, sensitivity of 81.82%, and specificity of 87.50%. The AUC of the HLR model was significantly higher than those of the LHR, RHR, and BHR models ( p < 0.001, DeLong test). DCA demonstrated superior net clinical benefit for the HLR model across relevant thresholds, while IDI and NRI confirmed significant improvement over the BHR model ( p&lt; 0.01). Wavelet-filtered GLSZM and GLRLM asymmetry metrics were identified as primary predictors. Conclusions: Hippocampal texture asymmetry features may serve as feasible imaging biomarkers for differentiating anti-LGI1 encephalitis from anti-GABABR encephalitis.