This material is presented to ensure timely dissemination of scholarly and technical work. Copyright and all rights therein are retained by authors or by other copyright holders. All persons copying this information are expected to adhere to the terms and constraints invoked by each author's copyright.
ICAI 2025 Posters, Nanning, China, November 6-9, 2025
- ProAttUnet: Protein Secondary-Structure Prediction Re-imagined via ESM2-Enhanced U-Net Dual-Fusion, ICAI 2025 Posters, Nanning, China, November 6-9, 2025
Authors: Long Cheng, Zhiqiang Hui, and Anchi Sun
Abstract: Protein secondary structure prediction remains a pivotal concern within the domain of bioinformatics. In this innovative research, we introduce a novel methodology to further enhance a protein prediction model grounded in single sequences. Our key contribution lies in integrating the state-of-the-art (SOTA) model ESM2, which hails from the field of universal protein language models. By leveraging ESM2, we are able to acquire residual embeddings and contact maps for the protein sequences under study. Regarding the model architecture, we employ a unique dual-way U-Net framework for effective feature fusion. This framework is complemented by the integration of a cross-attention mechanism, enabling the model to capture more comprehensive context information. Furthermore, In accordance with the distinctive characteristics of protein sequences, we incorporate a so-called GCU_SE module into both the encoder and the decoder components of the model. These innovative enhancements enable the ProAttUnet model to outperform the benchmark model SPOT-1D-Single by 1.6%, 3.5%, 1.0%, 4.6%, and 7.2% for ss3, and by 5.5%, 7.8%, 4.1%, 8.1%, and 10.1% for ss8 across five test sets (SPOT-2016, SPOT-2016-HQ, SPOT-2018, SPOT-2018-HQ and TEST2018, respectively). This significant improvement vividly demonstrates the effectiveness and novelty of our proposed model.
Keyword: Protein Secondary Structure Prediction, Protein Language Model, Cross Attention Mechanism, U-Net Architecture, Dynamic Sliding Window
DOI: 10.65286/icai.v3i1.12018
Cite - Optimization and Validation of Transformer-Based PTM Site Prediction Model for Paeonia lactiflora, ICAI 2025 Posters, Nanning, China, November 6-9, 2025
Authors: Kai Xiao, Wenzheng Bao
Abstract: Post-translational modifications (PTMs) significantly regulate peony’s growth, stress resistance, and biosynthesis of active pharmaceutical ingredients. However, traditional experimental methods for peony PTM site identification are cumbersome and inefficient; existing computational models are further limited by reliance on manual features, single modification type support, and poor interpretability—failing to meet precise identification needs.To address this, we first built a peony PTM site dataset: retrieving peony proteins from TCMSP, truncating sequences via sliding window to generate 1080 positive samples and 1976 negative sample, with sequence lengths of 3–41 amino acid residues.We then used the Transformer model for prediction: it fuses word vectors and position vectors for initial sequence representation, while its multi-head self-attention captures long-range residue interactions to explore PTM site patterns.10-fold cross-validation showed optimal performance at a sliding window length of 31; key metrics (accuracy, MCC, F1) significantly outperformed existing models, validating the approach’s effectiveness for peony PTM site identification.
Keyword: Transformer, PTM,Paeonia lactiflora, Classification
DOI: 10.65286/icai.v3i1.56688
Cite - Prediction of Post-Translational Modification Sites of Paeonia lactiflora Proteins Based on Attention Mechanism, ICAI 2025 Posters, Nanning, China, November 6-9, 2025
Authors: Yingyue Tang, Wenzheng Bao
Abstract: The precise prediction of post-translational modification sites of proteins was an important aspect of elucidating the mechanisms of traditional Chinese medicine. Existing computational models struggled to meet the needs of herbal pharmacological research due to their neglect of the specificity of herbal materials and the complexity of modification sites. This study focused on the traditional Chinese medicinal material Paeonia lactiflora and constructed a deep learning model based on convolutional neural networks and attention mechanisms by integrating multi-dimensional features of amino acid sequences. A dataset for Paeonia lactiflora was built using the TCMSP database, which included 1080 positive samples and 1976 negative samples, and the input space was optimized through feature normalization and dimensionality reduction. Experimental results indicated that the model effectively captured the modification patterns of proteins from different herbal materials, and SHAP feature selection significantly improved the prediction accuracy of post-translational modification sites. Compared to traditional single algorithm models, the proposed integrated architecture demonstrated significant advantages in balancing sequence conservation and functional specificity, providing a new computational tool for elucidating the mechanisms of action of active components in traditional Chinese medicinal materials and guiding rational clinical medication.
Keyword: Post-translational Modification Sites, Attention Mechanism, Transfer Learning, Paeonia lactiflora
DOI: 10.65286/icai.v3i1.67535
Cite
公安部备案号：