Alternative splicing generates multiple isoforms from a single gene. This process increases the diversity of gene functions as well as interactions with non-coding RNAs. Although gene functions and interactions have been studied extensively, little is known about isoform functions and their interactions with non-coding RNAs. In this thesis, we first study the isoform function prediction problem. We propose a novel deep learning method, DeepIsoFun, that combines multiple instance learning with domain adaptation. The latter technique helps to transfer the knowledge of gene functions to the prediction of isoform functions and provides additional labeled training data. Our model is trained on a deep neural network architecture so that it can adapt to different expression distributions associated with different gene ontology terms. Next, we approach the problem of predicting interactions between long non-coding RNAs (lncRNAs) and protein isoforms. We propose a novel method, DeepLPI, that combines heterogeneous data using a hybrid framework by integrating a deep neural network and a conditional random field. To overcome the lack of known interactions between lncRNAs and protein isoforms, we adopt a multiple instance learning approach again. Finally, we propose a new deep learning method, RESmim, to predict interactions between microRNAs and isoforms. It adapts our previous framework but uses a residual neural network to explore various levels of the feature space. We test our methods on different organisms including human, mouse, arabidopsis, and fruit-fly. They perform significantly better than the existing methods in both cross-validation and de novo prediction experiments. The experimental results demonstrate that our methods are effective in identifying the diverse functions and interactions of isoforms.