Predicting　disease-RNA　associations　is　important　for　disease　diagnosis　and　treatment.　The　traditional　biological　experiment　method　has　the　disadvantage　of　being　time-consuming　and　laborious.　Therefore,　a　growing　number　of　studies　have　focused　on　predicting　disease-RNA　associations　using　deep　learning　methods.　Aiming　at　the　characteristics　of　sparse　disease-RNA　association　data　and　few　labels,　we　propose　a　novel　prediction　method　named　SCLDA　based　on　contrastive　self-supervised　learning.　This　is　a　new　attempt　in　the　field.　We　also　propose　a　data　augmentation　method　based　on　RNA　similarity.　We　tested　SCLDA　and　other　advanced　methods　on　lncRNA　and　miRNA　datasets,　respectively,　and　the　results　show　that　SCLDA　achieves　the　best　performance.　©　2022　IEEE.