<p>To address the challenge of reliable early identification of surrounding-rock deterioration and large deformation in deep-buried soft rock tunnels under high in-situ stress, this study proposes a deformation-calibrated stability assessment and early-warning method based on microseismic (MS) monitoring and weighted time-varying networks. The results demonstrate that the time-varying MS event network can effectively capture the spatiotemporal evolution of MS activity, while the weighted network formulation further highlights structural evolution dominated by high-energy events. Under the spatiotemporal proximity thresholds of (<i>d</i>_<i>c</i>, <i>τ</i>_<i>c</i>)=(10 m, 8 h), the contributions of largest connected component ratio (<i>LCC</i>_<i>k</i>), average shortest path length (<i>L</i>_<i>k</i>), average clustering coefficient (<i>C</i>_<i>k</i>), and modularity (<i>Q</i>_<i>k</i>) to comprehensive stability index (<i>S</i>_<i>k</i>) were relatively balanced, accounting for 9.3%, 33.6%, 24.2%, and 32.9%, respectively. This parameter combination provided a more informative representation of the structural evolution of MS events. Across multiple tunnel sections, the multivariate logistic model achieved a high mean accuracy of 92%, a low mean false alarm rate of 7%, and strong discriminative capability, with a mean area under the curve (AUC) score of 0.94. The model enabled prediction of whether large deformation would occur within the subsequent 3 days, with a lead time of approximately 2 days. The proposed method provides a new technical pathway for the early warning of large deformation in similar projects.</p>