Simultaneous detection of influenza and SARS-CoV-2 on an AI-nanopore multiplex platform.

Seasonal influenza and SARS-CoV-2 co-circulate and present overlapping symptoms, creating demand for rapid multiplex diagnostics beyond the sensitivity limits of antigen tests and the infrastructure requirements of nucleic acid amplification tests. Here we report NanoMux, an amplification-free diagnostic platform that couples a solid-state nanopore with attention-based multiple-instance learning (Attention-MIL) to classify clinical saliva samples from their nanopore event streams. Each measurement is treated as a "bag" of unlabeled translocation events dominated by endogenous nanoparticles, while only a sparse subset contains virus like signatures. Attention-MIL learns to weight informative events during bag-level prediction, enabling robust classification in highly contaminated clinical matrices where conventional machine learning baselines degrade. In a three-class setting (SARS-CoV-2, influenza, and healthy controls), Attention-MIL achieved a macro F1 of 0.781, substantially outperforming a gradient-boosting baseline (0.327). Under diluted conditions, the LightGBM model exhibited a more gradual decline in accuracy than the Attention-MIL model, demonstrating greater robustness for low-concentration specimens. NanoMux establishes a physics-based, reagent-minimal route to rapid multiplex respiratory virus screening and provides a model framework for extracting sparse pathogen signals from noisy nanopore recordings.