Abstract: Antibodies are widely used in the prevention, diagnosis and treatment of various diseases. However, the success rate of therapeutic antibody development is far from satisfying. Many antibodies failed due to developability problems such as poor stability, low solubility, and cross-interactions or self-interactions. Whether a candidate monoclonal antibody is developable is closely related to its physicochemical properties. Although a few experimental assays are available to detect several types of physicochemical properties of antibody relevant to cross-interactions or self-interactions, they are laborious, time-consuming and expensive. Some computational methods for antibody developability evaluation have been reported. However, these methods are slow, low throughput, not available, too expensive, or not robust enough. In this paper, a support vector machine (SVM) model for predicting cross-interaction or self-interaction of antibodies is constructed by using dipeptide deviation from expected mean derived from antibody sequences as features. The ensemble model achieves 100% sensitivity and 96.18% accuracy in cross-validation. The model can be used for high-throughput assessment of cross-interaction or self-interaction of antibodies, speeding therapeutic antibodies development, and reducing cost. Based on the model, a free web server called CISI2.0 is built, which is available at http://i.uestc.edu.cn/CISI2.