Several approaches to learning probabilistic logic programs have been proposed in the literature. However, most learning systems based on these approaches are not efficient for handling large practical problems (especially, in the case of structure learning). It has been a challenging issue to reduce the search space of candidate (probabilistic) logic programs. There is no exception for SLIPCOVER, a latest system for both parameter and structure learning of Logic Programs with Annotated Disjunction (LPADs). This paper presents a new algorithm T-LPAD for structure learning of LPADs by employing transfer learning. The new algorithm has been implemented and our experimental results show that T-LPAD outperforms SLIPCOVER (and SLIPCASE) for most benchmarks used in related systems.