Relation-aware subgraph embedding is promising for predicting multi-relational drug-drug interactions (DDIs). Typically, most existing methods begin by constructing a multi-relational DDI graph and then learning relation-aware subgraph embeddings (RaSEs) of drugs from the DDI graph. However, most existing approaches are usually limited in learning RaSEs of new drugs, leading to serious over-fitting when the test DDIs involve such drugs. To alleviate this issue, We propose a novel DDI prediction method based on relation-aware subgraph embedding with co-contrastive learning, RaSECo. RaSECo constructs two heterogeneous drug graphs: a multi-relational DDI graph and a multi-attributes-based drug-drug similarity (DDS) graph. The two graphs are used respectively for learning and propagating the RaSEs of drugs, thereby ensuring that all drugs, including new ones, can aggregate effective RaSEs. Additionally, we employ a cross-view contrastive mechanism to enhance drug-pair (DP) embedding. RaSECo learns DP embeddings from two distinct views (interaction and similarity views) and encourages these views to supervise each other collaboratively to obtain more discriminative DP embeddings. We evaluate the effectiveness of our RaSECo on three different tasks using two real datasets. The experimental results demonstrate that RaSECo outperforms existing state-of-the-art prediction methods.