Discovering human disease-causing genes (disease genes in short) is one of the most challenging problems in bioinformatics and biomedicine, as most diseases are re- lated in some way to our genes. Various methods have been proposed to exploit existing data sources for solving the problem. We aim to develop a novel method to predict dis- ease genes that takes into account the imbalance between known disease genes and unknown disease genes. To this end, our method makes the best of semi-supervised learn- ing, integrating data of human protein-protein interactions and various biological data extracted from multiple pro- teomic/genomic databases. Experimental evaluation shows high performance of our proposed method. Also, a consid- erable number of potential disease genes were discovered. Supplementary materials are now available from http://www.jaist.ac.jp/ s0560205/DiseaseGenes/.