NAD_MCNN: Combining Protein Language Models and Multiwindow Convolutional Neural Networks for Deacetylase NAD+ Binding Site Prediction

Abstract

Sirtuins, a class of NAD+ -dependent deacetylases, play a key role in aging, metabolism, and longevity. Their interaction with NAD+ at the catalytic site is crucial for function, but experimental methods to map NAD+ binding sites are time consuming. To address this, we developed a computational method integrating pretrained protein language models with multiwindow convolutional neural networks (CNNs). This method captures sequence information and diverse local patterns, achieving state-of-the-art performance, with AUC of 0.9733 for human sirtuin proteins and 0.9701 for other NAD-dependent deacylation enzymes. These findings offer insights into the role of sirtuins in aging and their broader biological functions while providing a new path for identifying therapeutic targets in aging-related diseases.