OmiImp: A Cross-Omics Imputation Framework Based on Improved Generative Adversarial Network

The integration of multi-omics data has emerged as a powerful approach to elucidate interactions across different biological levels. However, throughput limitations and high costs of sequencing technologies often result in sparse multi-omics datasets, where only a subset of samples contains complete omics profiles - a challenge known as the “block missing”. To address this challenge, we propose OmiImp, a novel computational framework based on an improved generative adversarial network (GAN) for cross-omics data imputation. Through performance evaluation on independent datasets, we demonstrate that OmiImp outperforms existing state-of-the-art imputation methods while maintaining stable performance across different missing rates. In addition, we perform enrichment analysis and the results demonstrate that differential expressed features are uniformly distributed across pathways, and the synthetic data retains utility in diverse prognostic analyses. Collectively, this methodological advancement facilitates more reliable multi-omics integration studies, particularly when handling incomplete datasets.