Abstract: Cellular biological processes are temporally dynamic, and protein functional modules are the functional units that drive cellular biological processes. In order to identify protein functional modules, cellular biological processes were modelled as dynamically and temporally gene expression-associated protein-protein interaction networks（DTEPIN）. A sub-block matrix was constructed to represent DTEPIN. By employing the particularity of the sub-block matrix and analyzing time-space complexity and parallelism, Markov clustering algorithm was optimally designed to identify the protein functional modules in DTEPIN. In order to carry out the process of Markov clustering based on sub-block matrix, matrix multiplication using graphics processor unit was implemented to calculate matrix product in parallel. Experimental results show that compared with the existing similar algorithms, the designed algorithm can accurately identify more protein functional modules and identify more protein functional modules with higher quality.