Background: ROS1 rearrangement was found in about 1% non-small cell lung cancer (NSCLC), and the fusion proteins created during this process will result in the successive activation of the ROS1 kinase domain. NSCLC patients with ROS1 gene fusions were considered a unique subtype highly sensitive to related tyrosine kinase inhibitor (TKI) treatments. However, the acquired TKI resistance was the major hurdle preventing patients from getting prolonged treatment benefits. Methods: 107 patients diagnosed with advanced or metastatic NSCLC harboring ROS1 fusion were retrospectively recruited. All patients were treated with crizotinib as first-line treatment, and 21 patients received lorlatinib after crizotinib progression. Samples were collected at baseline, after crizotinib progression, and after lorlatinib progression, which all underwent targeted DNA sequencing. TKIs binding to mutated ROS1 fusion proteins was simulated using molecular dynamics simulations. Results: The most witnessed fusion partner of ROS1 was CD74 (58%), followed by SDC4 (14%), EZR (11%), and SLC34A2 (9%). The median progression-free survival was 12.9 months for crizotinib and 6.4 months for lorlatinib. Patients with CD74-ROS1 and SLC34A2-ROS1 had significantly longer PFS than those with other ROS1 fusion types when treated with crizotinib. Patients with baseline TP53 mutations showed worse PFS compared to TP53 wild-type (WT) patients (P < 0.001, HR: 0.19, 95%CI: 0.09-0.39) under crizotinib treatment. An accumulation of both on-target (baseline vs. post-crizotinib vs. post-lorlatinib: 0% vs. 43% vs. 62%) and off-target resistant mutations (baseline vs. post-crizotinib vs. post-lorlatinib: 22% vs. 26% vs. 43%) after multiple TKI treatments was observed. A total of ten on-target resistance mutations were detected after TKI therapies, with 4 first-reported mutations (ROS1 L2010M, G1957A, D1988N, L1982V). According to the molecular dynamics simulation results, ROS1 L2010M mutations (happening in the ROS1 kinase ATP binding cassette) maybe resistant to lorlatinib, entrectinib, cabozantinib, and crizotinib. ROS1 G1957A may lead to resistance to cabozantinib. Moreover, all four novel on-target mutations may lead to resistance to crizotinib. Conclusions: In summary, CD74- and SLC34A2-ROS1 patients showed better crizotinib efficacy with different resistance mutation patterns. Patients of these subtypes are potential beneficiaries of molecular testing after crizotinib progression, directing future treatment strategies. Multiple TKI treatments may lead to the accumulation of both on-target and off-target resistance mutations. In addition, 4 novel ROS1 on-target mutations identified underwent molecular dynamics simulations, unveiling potential resistance to different TKIs, which can provide vital information for future ROS1 fusion-positive NSCLC patient treatment selections.