Abstract
Moonlighting proteins perform multiple, distinct biological functions beyond their primary role, posing challenges in traditional protein function annotation. In this study, we present an artificial intelligence-based framework to identify moonlighting proteins, employing advanced machine learning models integrated with Local Interpretable Model-agnostic Explanations. Local Interpretable Model-agnostic Explanations provides interpretability by highlighting the features driving the model’s predictions, bridging the gap between performance and explainability. Our results demonstrate 92% accuracy in distinguishing moonlighting proteins from non-moonlighting proteins and 97.6% area under the curve, validated against benchmark datasets. Furthermore, Local Interpretable Model-agnostic Explanations (LIME) explanations reveal biologically plausible insights, such as domain-level correlations and structural motifs associated with multifunctionality. This interpretability not only enhances trust in artificial intelligence predictions but also offers novel hypotheses for experimental validation. This work signifies a step toward transparent and reliable artificial intelligence applications in computational biology.
Fatemeh Zare-Mirakabad
Associate Professor
My research interests include bioinformatics, computational biology and artificial intelligence.