MIF ‐ MAPMS : Enhancing identification of myelin autoantigenic peptides in multiple sclerosis through multimodal information fusion

Information from the abstract

Multiple sclerosis (MS) arises from an autoimmune response in which the immune system erroneously targets myelin autoantigens within the central nervous system, leading to myelin degradation and subsequent neurological dysfunction. Identifying myelin autoantigenic peptides (MAPs) is therefore critical for understanding MS pathogenesis and developing targeted therapies; however, conventional experimental approaches remain time-consuming and costly. Thus, computational methods that can perform in silico screening of T cell-specific MAP in MS (MAPMSs) using only peptide sequences are highly desirable. Existing computational methods primarily rely on a single modality, which often fails to capture key information of MAPMSs, leading to limited sequence representation and generalization ability. To address this limitation, we propose MIF-MAPMS, a novel multimodal information fusion framework that leverages multimodal information, including peptide format and SMILEs notation, for accurate MAPMS identification. This novel framework processes different modalities of compositional descriptors, molecular fingerprints, ESM-2 embeddings, and Mol2V embeddings using specific deep learning methods, leading to enriched MAPMS representation. Subsequently, the extracted embeddings are fused and passed through a multilayer perceptron (MLP), followed by a fully connected neural network for MAPMS identification. Both cross-validation and independent test results show that MIF-MAPMS attains significant improvements in MAPMS identification over the benchmark main and alternative datasets, with Matthew's correlation coefficient (MCC) of 0.931-0.968 and 0.812-0.928, providing 5.78%-8.04% and 1.22%-2.98% increases, respectively, compared to the existing method. Ablation studies further confirm the necessity of multimodal information fusion in improving MAPMS representation and the model's predictive performance. All codes and datasets are freely available online at https://github.com/lawankorn-m/MIF-MAPMS.

Why this record is monitored

This record has an Impact Signal of 72/100 based on recency, source, collaboration, and bibliographic signals. It prioritizes monitoring and is not a judgment of research quality.

Related topics: vaccines and immunoinformatics approaches · Machine Learning in Bioinformatics · Fractal and DNA sequence analysis