TiPo: Pretrained Models and Inference Resources for Bacterial Functional Identification

This repository provides the pretrained model weights and associated inference resources for TiPo, an interpretable deep learning framework for bacterial functional identification and key functional gene discovery. TiPo integrates 3–7-mer frequency features with weighted feature representation, cross and residual structures, deep neural network stacking, and a TransformerEncoder architecture incorporating an Agent-Attention mechanism. The framework is designed for the integrated identification of beneficial bacteria and bacterial pathogens and supports model interpretation for the discovery of informative genomic features and key functional genes. The files deposited here provide the pretrained models and associated preprocessing resources required to reproduce TiPo inference. Source code, usage instructions, and software dependencies are maintained separately in the accompanying GitHub repository.