[Proceeding] Graph-affiliated Unsupervised Segmentation Assisted Simple Neural Network

High-throughput 16S rRNA-seq data, complemented by detailed cultivation information, constitutes a critical resource in bacterial research with promising implications for biomedical applications such as fecal microbiota transplantation. However, the inherent high dimensionality and substantial costs associated with 16S rRNA-seq data constrain its full utility. In this paper, we introduce a novel approach—graph-affiliated unsupervised segmentation-assisted simple neural network (GASNN)—designed to analyze 16S rRNA-seq data efficiently. In a proof-of-concept application involving the prediction of cultivation media temperature, the GASNN model achieved significant performance enhancements over a traditional simple neural network (SNN). Further experiments across various tasks consistently demonstrated that GASNN improves the performance of SNN models. Nevertheless, a notable limitation of the proposed approach is that its benefits may diminish as the network architecture deepens, thereby impeding its ability to reveal the intrinsic manifold structure of the data.

Recommended citation: [TBD] Juntang Wang, Runkun Guo, Dongmian Zou, Shixin Xu (2025). "Graph-affiliated Unsupervised Segmentation Assisted Simple Neural Network." 2025 IEEE International Conference on Bioinformatics and Biomedicine (BIBM). 1(1).
Download Paper