This review summarizes recent deep learning-based approaches for multi-omics data integration and analysis. The field of multi-omics data integration has seen significant progress due to the rapid development of deep learning and the increasing availability of complex and heterogeneous data. Multi-omics data includes molecular and imaging modalities, which, when combined, can improve performance in tasks such as prediction, classification, and clustering. Deep learning encompasses a wide variety of methods, each with unique strengths and weaknesses for multi-omics integration. The review categorizes deep learning-based methods into non-generative and generative approaches. Non-generative methods include feedforward neural networks, graph convolutional neural networks, and autoencoders. These methods focus on learning the conditional probability distribution of the outcome given the input, rather than the joint probability distribution of the input and labels. Generative methods, such as variational autoencoders, generative adversarial networks, and generative pretrained transformers, can impose constraints on the shared representations to enforce certain properties or incorporate prior knowledge. They can also be used to generate or impute missing modalities. The review discusses several emerging themes in multi-omics integration, including the use of generative methods and attention mechanisms. It also highlights the importance of handling missing data and integrating more data types to improve performance on downstream tasks. The review introduces a distinctive approach by incorporating generative pretrained transformers (GPT), which previous studies have not extensively utilized. From an application perspective, the review addresses the challenge of incomplete data and broadens the scope to include imaging modalities. The review also discusses the advantages and limitations of different methods, including their ability to handle missing data, their computational complexity, and their suitability for different types of data. Overall, the review emphasizes the importance of integrating multiple data types and using deep learning methods that can handle missingness and provide interpretable results. The review concludes that further growth in methods that can handle missingness and integrate more data types is expected, as these are common challenges in working with complex and heterogeneous data.This review summarizes recent deep learning-based approaches for multi-omics data integration and analysis. The field of multi-omics data integration has seen significant progress due to the rapid development of deep learning and the increasing availability of complex and heterogeneous data. Multi-omics data includes molecular and imaging modalities, which, when combined, can improve performance in tasks such as prediction, classification, and clustering. Deep learning encompasses a wide variety of methods, each with unique strengths and weaknesses for multi-omics integration. The review categorizes deep learning-based methods into non-generative and generative approaches. Non-generative methods include feedforward neural networks, graph convolutional neural networks, and autoencoders. These methods focus on learning the conditional probability distribution of the outcome given the input, rather than the joint probability distribution of the input and labels. Generative methods, such as variational autoencoders, generative adversarial networks, and generative pretrained transformers, can impose constraints on the shared representations to enforce certain properties or incorporate prior knowledge. They can also be used to generate or impute missing modalities. The review discusses several emerging themes in multi-omics integration, including the use of generative methods and attention mechanisms. It also highlights the importance of handling missing data and integrating more data types to improve performance on downstream tasks. The review introduces a distinctive approach by incorporating generative pretrained transformers (GPT), which previous studies have not extensively utilized. From an application perspective, the review addresses the challenge of incomplete data and broadens the scope to include imaging modalities. The review also discusses the advantages and limitations of different methods, including their ability to handle missing data, their computational complexity, and their suitability for different types of data. Overall, the review emphasizes the importance of integrating multiple data types and using deep learning methods that can handle missingness and provide interpretable results. The review concludes that further growth in methods that can handle missingness and integrate more data types is expected, as these are common challenges in working with complex and heterogeneous data.