The genome sequence of crop species has become an essential resource for advancing modern intelligent breeding techniques. However, many crop species, such as potatoes, are polyploids, and the high sequence similarity among homologous chromosomes makes it challenging to accurately reconstruct all the haplotypes of polyploid genomes. To address this challenge, we first design experiments leveraging the latest long-read sequencing, chromosome conformation capture sequencing, and single-cell sequencing technologies etc to generate high-throughput genomic data, and then we develop novel computational algorithms and tools for analyzing this extensive data and understanding the biological system.

Our lab is at the forefront of integrating artificial intelligence with single-cell multi-omics analysis. We have developed several innovative platforms and methodologies to enhance our understanding and interpretation of complex biological data. Our work includes the development of a spatial single-cell multi-omics analysis platform, which allows for comprehensive data collection and analysis. We also focus on multi-platform multi-source data integration and dimensionality reduction inference to construct homologous structure similarity frameworks.

Additionally, we have created a spatial single-cell transcriptome analysis platform, enabling precise data transformation and exploration. Our research extends to cross-platform data integration and multi-scale data transfer learning solutions, ensuring robust and scalable analytical frameworks. Furthermore, we investigate multi-modal data integration and non-linear inference for dimensionality reduction control, emphasizing the importance of combining various data types for more accurate results. Lastly, our efforts include multi-omics feature extraction and temporal convolutional network frameworks, highlighting our commitment to advancing theoretical models and framework research in this domain.

Our lab focuses on multi-modal (view) data processing, especially on multi-modal graph data and multi-source data with data privacy constrain under various Non-ideal data annotation circumstances. We have development several state-of-the-art methods to handle multi-modal (view) and weakly-labeled data feature selection, representation learning and clustering. Meanwhile, several foundation optimization methods, such as Re-weighted method, fast coordinate descent method, rank-based optimization method, etc, to handle the optimization problems involved in different machine learning cases. Recently, we design several multi-modal data fusion methods for single-cell multi-omics data, RNA 3D structure data, etc. The preprint will be shared soon.