MOTIVATION: Global population aging has led to a rapid increase in neurodegenerative disorders such as Alzheimer's disease (AD). Although existing drugs can temporarily alleviate symptoms, none have been proven to delay or prevent disease progression. Acetylcholinesterase inhibitors (AChEIs) have been shown to mitigate AD symptoms, yet traditional AChEI screening approaches remain time-consuming and inefficient. RESULTS: To address this limitation, we developed multi-species AChEI screening network (MAISNet), an AChEI screening framework based on acetylcholinesterase (AChE) data from six species. In MAISNet, inhibitor molecules were represented as SMILES-derived molecular graphs, whereas AChE protein structures were encoded as residue contact maps. Multi-scale molecular and protein features were extracted using the sample and aggregate (GraphSAGE) network and the graph attention network, respectively, and were subsequently fused through a bidirectional cross-attention mechanism. The integrated representations were then processed by a multilayer perceptron (MLP) to inhibitor classification. On both internal and external validation sets, MAISNet consistently outperformed five baseline models. Furthermore, we applied MAISNet to screen existing small molecules, and Methyl 2-[(3S)-3-(1, 2, 3, 4, 5, 6, 7, 8-octahydro-2-naphthyl)-2-(methoxycarbonyl)-1H-pyrrol-1-yl]acetate subsequently emerged as the top-ranked candidate. Overall, MAISNet significantly improves the accuracy and generalization capability of AChEI screening, providing an efficient and reliable computational tool for accelerating therapeutic discovery for AD. AVAILABILITY AND IMPLEMENTATION: Code that supports the reported results can be found at: https://github.com/liangshengjie111/MAISNet. The archival version of the code is preserved on Zenodo at https://doi.org/10.5281/zenodo.18721665.

Identifying aggregation-prone proteins or peptides is essential for advancing our understanding of amyloid aggregation processes and their related pathogenic mechanisms. Recognizing potential amyloid hexapeptides can also support peptide-based drug design and reduce experimental costs. In this study, we proposed TSLAmy, a computational model designed to predict amyloid hexapeptides using a two-stage learning framework. In the first stage, we performed feature extraction on the hexapeptides, and in the second stage, we presented prediction model for amyloid hexapeptide aggregation. Firstly, to ensure balanced dataset partitioning, we applied a clustering-based method by training two autoencoders on all possible hexapeptides using their sequence and physicochemical features, respectively. The resulting clusters were used to stratify the data into training and testing datasets. Then, in the first stage, we extracted features from hexapeptides based on their sequence and physicochemical properties. The feature extraction module was used to obtain physicochemical features, while the ESM-2 module was responsible for extracting sequence features for each hexapeptide. Finally, in the second stage, the aggregation prediction module was employed to predict the aggregation potential of hexapeptides. The experimental results demonstrated that the accuracy of TSLAmy reached 0.8493 (0.8447-0.8539), outperforming other state-of-the-art methods. Furthermore, we predicted the aggregation potential of all 64,000,000 possible hexapeptides and analyzed the amino acids that form aggregation-prone hexapeptides. We anticipate that TSLAmy can offer new insights into the identification of aggregation-prone peptides, contributing to advancements in peptide drug development.

Bronchoalveolar lavage fluid (BALF) is a liquid obtained from the alveoli and bronchi, often used to study pulmonary diseases. So far, proteomic analyses have identified over three thousand proteins in BALF. However, the comprehensive characterization of these proteins remains challenging due to their complexity and technological limitations. This paper presented a novel deep learning framework called SecProGNN, designed to predict secretory proteins in BALF. Firstly, SecProGNN represented proteins as graph-structured data, with amino acids connected based on their interactions. Then, these graphs were processed through graph neural networks (GNNs) model to extract graph features. Finally, the extracted feature vectors were fed into a multi-layer perceptron (MLP) module to predict BALF secreted proteins. Additionally, by utilizing SecProGNN, we investigated potential biomarkers for lung adenocarcinoma and identified 16 promising candidates that may be secreted into BALF.