Predicting Ampicillin Resistance in Escherichia Coli using an Adaptive Neuro-Fuzzy Inference System (ANFIS): A Proof-of-Concept Computational Approach for Phenotypic Susceptibility Modeling

Abstract

The increasing prevalence of antimicrobial resistance (AMR) in Escherichia coli threatens the effectiveness of β-lactam antibiotics in both veterinary and human medicine. Conventional susceptibility testing provides categorical interpretations of bacterial response, whereas computational intelligence approaches can model resistance as a continuous and potentially non-linear biological phenomenon. A first-order Takagi-Sugeno Adaptive Neuro-Fuzzy Inference System (ANFIS) was developed to predict ampicillin (AMP, 10 μg) inhibition-zone diameters of bovine E. coli isolates. The model architecture consisted of 78 nodes, 27 fuzzy inference rules, and 54 adjustable parameters. Model training was performed using seven experimentally derived susceptibility observations through a hybrid learning algorithm combining least-squares estimation and gradient descent optimization.The ANFIS model achieved a minimum training Root Mean Square Error (RMSE) of 0.308607, indicating close agreement between observed and predicted values within the training dataset. The predicted inhibition-zone diameter for the evaluated isolate was 11.89 mm, corresponding to a resistant phenotype according to CLSI interpretive criteria (≤ 13 mm). The model successfully generated a continuous susceptibility estimate rather thana simple categorical classification, thereby preserving quantitative information related to bacterial response to ampicillin.The findings demonstrate the feasibility of applying neuro-fuzzy inference systems to antimicrobial susceptibility modeling. Nevertheless, the marked disparity between the number of trainable parameters and available observations (54 versus 7) suggests a substantial risk of overfitting and limits the capacity for biological generalization.Consequently, the present model should be regarded as a proof-of-concept framework requiring validation using larger datasets, cross-validation procedures, and independent external testing before practical deployment.