Predicting drug-disease interactions is becoming essential for thoroughly recognition drugs’ functions and mechanisms. However, it is a laborious, costly and time-consuming progress for approving a new drug to treat a disease. Therefore, developing computational approaches for uncovering drug-disease interactions is attracting researchers from various fields. In this study, a model entitled CFDSAEDDA is introduced for inferring drug-disease associations. Firstly it integrates multiple similarities to form new integrated similarities. Secondly, the problem of sparse drug-disease interactions is solved by utilizing a collaborative filtering algorithm. Finally, a deep sparse autoencoder technique, which can handle complex data, is employed to provide high quality prediction results. CFDSAEDDA achieves a curious performance in inferring drug-disease associations demonstrated by the values 0.9685 and 0.9864 of AUC and AUPR, respectively, under 10-fold cross validation experiments on the Cdataset. It is competitive to related methods when comparing on the same Cdataset. Thus, it could be acknowledged to be a practical tool for unveiling drug-disease relationships.