We demonstrate an automatic, high-throughput and high-sensitivity particle aggregation-based sensor that uses wide-field, compact and cost-effective lens-less microscopy, powered by deep neural networks. In this method, the post-reaction assay is imaged by a snapshot hologram over a wide field-of-view (20mm²). Using a deep learning-based holographic reconstruction, all the particle clusters are simultaneously reconstructed in ~30s. Using this method, we demonstrated accurate and rapid readout of an immunoassay to detect herpes simplex virus, which affects >50% of the adults in US, and achieved a clinically-relevant detection limit (~ 5viruses/µL). This method can be broadly used to quantify other particle-aggregation based immunoassays.