deepAMPNet: a novel antimicrobial peptide predictor employing AlphaFold2 predicted structures and a bi-directional long short-term memory protein language model

PeerJ, Год журнала: 2024, Номер 12, С. e17729 - e17729

Опубликована: Июль 19, 2024

Background Global public health is seriously threatened by the escalating issue of antimicrobial resistance (AMR). Antimicrobial peptides (AMPs), pivotal components innate immune system, have emerged as a potent solution to AMR due their therapeutic potential. Employing computational methodologies for prompt recognition these indeed unlocks fresh perspectives, thereby potentially revolutionizing drug development. Methods In this study, we developed model named deepAMPNet. This model, which leverages graph neural networks, excels at swift identification AMPs. It employs structures predicted AlphaFold2, encodes residue-level features through bi-directional long short-term memory (Bi-LSTM) protein language and constructs adjacency matrices anchored on amino acids’ contact maps. Results comparative study with other state-of-the-art AMP predictors two external independent test datasets, deepAMPNet outperformed in accuracy. Furthermore, terms commonly accepted evaluation such AUC, Mcc, sensitivity, specificity, achieved highest or highly comparable performances against predictors. Conclusion interweaves both structural sequence information AMPs, stands high-performance that propels evolution design peptide pharmaceuticals. The data code utilized can be accessed https://github.com/Iseeu233/deepAMPNet .

Язык: Английский

---

PAPreC: A Pipeline for Antigenicity Prediction Comparison Methods across Bacteria

ACS Omega, Год журнала: 2025, Номер 10(6), С. 5415 - 5429

Опубликована: Фев. 3, 2025

Antigenicity prediction plays a crucial role in vaccine development, antibody-based therapies, and diagnostic assays, as this predictive approach helps assess the potential of molecular structures to induce recruit immune cells drive antibody production. Several existing methods, which target complete proteins epitopes identified through reverse vaccinology, face limitations regarding input data constraints, feature extraction strategies, insufficient flexibility for model evaluation interpretation. This work presents PAPreC (Pipeline Prediction Comparison), an open-source, versatile workflow (available at https://github.com/YasCoMa/paprec_nx_workflow) designed address these challenges. systematically examines three key factors: selection training sets, methods (including physicochemical descriptors ESM-2 encoder-derived embeddings), diverse classifiers. It provides automated evaluation, interpretability SHapley Additive exPlanations (SHAP) analysis, applicability domain assessments, enabling researchers identify optimal configurations their specific sets. Applying IEDB reference, we demonstrate its effectiveness across ESKAPE pathogen group. A case study involving Pseudomonas aeruginosa Staphylococcus aureus shows that are more suitable different sequence types, embeddings enhance performance. Moreover, our results indicate separate models Gram-positive Gram-negative bacteria not required. offers comprehensive, adaptable, robust framework streamline improve antigenicity bacterial

Язык: Английский

---

Artificial intelligence-powered prediction of AIM-2 inflammasome sequences using transformers and graph attention networks in periodontal inflammation

Scientific Reports, Год журнала: 2025, Номер 15(1)

Опубликована: Март 13, 2025

Periodontal inflammation is a chronic condition affecting the tissues surrounding teeth. Initiated by dental plaque, it triggers an immune response leading to tissue destruction. The AIM-2 inflammasome regulates this response, and understanding its peptide sequences could aid in developing targeted therapeutics. This study explores using transformers graph attention networks (GAT) treat periodontal inflammation. UniProt was used download proteins FASTA with 100%, 90%, 50% similarity. DeepBio, web service for deep-learning architectures, analyzed these sequences. Peptide sequence prediction methods were evaluated transformer, RNN-CNN, GAT models. transformer model achieved 84% accuracy, 86%, RNN-CNN 64%. Both models predicted more effectively than model, Transformer showing highest class accuracy at 85%, followed 80%. Models exhibited varying sensitivity specificity, demonstrating superior performance overall class-specific prediction. AI-based transformers, GAT, shows promise accurately predicting sequences, outperforming accuracy.

Язык: Английский

---

iPiDA-LGE: a local and global graph ensemble learning framework for identifying piRNA-disease associations

BMC Biology, Год журнала: 2025, Номер 23(1)

Опубликована: Май 9, 2025

Abstract Background Exploring piRNA-disease associations can help discover candidate diagnostic or prognostic biomarkers and therapeutic targets. Several computational methods have been presented for identifying between piRNAs diseases. However, the existing encounter challenges such as over-smoothing in feature learning overlooking specific local proximity relationships, resulting limited representation of pairs insufficient detection association patterns. Results In this study, we propose a novel method called iPiDA-LGE identification. comprises two graph convolutional neural network modules based on global graphs, aimed at capturing general features pairs. Additionally, it integrates their refined macroscopic inferences to derive final prediction result. Conclusions The experimental results show that effectively leverages advantages both learning, thereby achieving more discriminative pair superior predictive performance.

Язык: Английский

---

Inter- and intra-uncertainty based feature aggregation model for semi-supervised histopathology image segmentation

Expert Systems with Applications, Год журнала: 2023, Номер 238, С. 122093 - 122093

Опубликована: Окт. 14, 2023

Язык: Английский

---