Based on the new drug research and development (R&D) cooperation data of different stages from 2015 to 2021, this study explored the heterogeneity in the composition of entities and spatial structure across different stages of new drug research and development by using social network analysis, negative binomial models and other methods. It identified various types of collaborations in new drug research and development and further discussed the main factors driving such collaborations in China. The main results are as follows: firstly, there are differences in actor composition across various stages of new drug R&D. Universities and research institutions are key actors in the pre-clinical trial phase, whereas enterprises play a dominant role during the clinical trial phase. Secondly, overall, the topological structure of China's new drug R&D collaboration network has evolved from a "dual-core" driven pattern to a "dual-core+multi-polar" driven pattern. The "dual-core" refers to Beijing and Shanghai, while the "multi-polar" nodes include cities such as Suzhou, Guangzhou, Chengdu, and Nanjing. Moreover, the spatial configuration of the collaboration network differs significantly between the pre-clinical and clinical trial phases. Thirdly, in the pre-clinical trial phase, urban R&D collaborations tend to shift from internal to external partnerships, whereas the direction of this shift is reversed during the clinical trial phase. Fourthly, geographical proximity plays a significant role in all stages of new drug R&D collaboration, while factors such as technological and institutional proximity exhibit certain phase-specific and stage-dependent variations.