This paper presents DDGrid, a novel Grid computing system for drug discovery and design. By utilizing the idle resources donated by the clusters that scatter over the Internet, DDGrid can implement efficient data-intensive biologic applications. The high-level resource management framework with a Grid-P2P hybrid architecture is described. With P2P technologies, some problems which are inevitable in the master-slave model can be avoided, such as single point of failure or performance bottleneck. Then an agent-based resource scheduling algorithm is presented. With this scheduling algorithm, the idle computational resources are dynamically scheduled according to the real-time working load on each execution node. Thus DDGrid can hold an excellent load balance state. Furthermore, the framework is introduced into the practical protein molecules docking applications. Solid experimental results show the load balance and robustness of the proposed system, which can greatly speed up the process of protein molecules docking.