TransVi: A Transformer-based Web Application for Pathogenic Virus Identification

TransVi is a web-based classification model that leverages transformer architectures to predict pathogenic viruses directly from metagenomic sequences. To achieve this, the classification model is first pretrained with DNABERT on genomic sequences, capturing biological context and long range dependencies. The model is then fine-tuned with labelled viral data to enable high accuracy identification of strains such as SARS-CoV-1, MERS, SARS-CoV-2, Ebola, Dengue, and Influenza. TransVi demonstrates the power of large language models in resolving short read ambiguities and offers a scalable solution for real time pathogen surveillance and outbreak response.

MetaTrans: A Transformer-Integrated Web Application to Improve the Detection of Pathogenic Strains from Metagenomic Data

MetaTrans integrates both supervised and unsupervised techniques to improve strain level pathogen detection from metagenomic data. It employs large language models and operates across three phases: initial classification (Model: CLM), clustering of unlabeled data (Model: CLT), and retraining with enriched annotations (Model: CLM*). This iterative pipeline enhances sensitivity and contextual accuracy, enabling precise identification of viruses like SARS-CoV-1, MERS, and SARS-CoV-2. MetaTrans can support intelligent annotation for biomedical research and public health applications.