|Company:||Agency for Science, Technology and Research|
Artificial Intelligent (AI)/Bioinformatician research fellow (postdoctoral) positions are available in Jinmiao Chen's (CJM) lab. Characterization of immune cell heterogeneity and cell-cell interactions at single-cell resolution is critical for understanding the fundamental mechanisms used by immune cells to promote or prevent disease progression and response to treatment. Single-cell multi/spatial-omics technology, an ideal tool for immunology, is rapidly advancing and generating big, complex and heterogeneous datasets, which poses significant challenges on data analysis and integration, meanwhile also provides opportunities for the development of data-driven, bio-inspired artificial intelligence (AI). CJM lab combines single-cell multi/spatial-omics with AI for precision immunology research. Her research spans 3 themes ():
Theme 1: Build human single cell atlases by deep integration of public single-cell omics data. The ability to study cellular heterogeneity at single cell resolution is making single-cell sequencing increasingly popular. However, there is no publicly available resource that offers an integrated cell atlas with harmonized metadata that users can integrate new data with. We develop DISCO (), a database of Deeply Integrated Single-Cell Omics data. The current release of DISCO integrates more than 18 million cells from 4450 samples, covering 107 tissues / cell lines / organoids, 158 diseases, and 20 platforms. We standardized the associated metadata with a controlled vocabulary and ontology system. To allow large scale integration of single-cell data, we developed FastIntegrate which also returns batch corrected gene expression values. We also developed CELLiD, an atlas guided automatic cell type identification tool. Employing these two tools on the assembled data, we constructed one global atlas and 26 sub-atlases for different tissues, diseases, and cell types. As single cell technological developments continue apace, we intend to continuously update and upgrade DISCO. We will update DISCO as new studies are published. New sub-atlases will also be constructed and annotation updated as needed to reflect any new developments. We also plan to expand the scope of DISCO to encompass other single-cell omics data, such as scATAC-seq, scTCR-seq, scBCR-seq, and spatial transcriptomics. The different omics data will be integrated to create a single cell multi-omics reference atlas.
1.Database and web portal development
2.Data integration / batch effect correction with disentangled representation learning
3.Cell type identification with transfer learning
4.Continual learning that continuously integrates new data to be published
Theme 2: Develop and deploy deep multimodal representation learning methods for the analysis and integration of single-cell multi-omics. Single-cell multi-omics simultaneously measures multiple-omes of a cell including its genome, transcriptome, epigenome, methylome, proteome, immune receptor repertoires and etc, providing a holistic view of individual cells and holds an unprecedented potential for dissecting cellular heterogeneity. Compared to single-omics, multi-omics is able to identify subtler differences between cells and reveal links across omes. When analysing such data, combining the multiple-omes to learn a low-dimensional discriminative representation for each cell is crucial. Specifically, joint embeddings of the multiple data modalities are essential for various down-stream analysis tasks such as visualization, clustering, trajectory inference, and batch integration. CJM lab builds AI models to learn each data modality by dedicated deep neural network and then jointly train them with multi-view learning to produce an unsupervised embedded deep representation of cells. With this, we seek to achieve higher resolution of cell type identification discover new cell populations & their associated functions uncover relationships across-omics and predict across modalities.
1.Multimodal AI / multi-view learning
Theme 3: Build human spatial omics atlas and construct cell-cell interactomes.
In current single cell analysis, cell-cell interactions are computationally predicted solely based on ligand and receptor expression and without spatial information. The emerging spatial omics technologies enable simultaneous measurement of gene/protein expression and cell locations, which is critical for characterizing cell-cell interactions. We first build human spatial omics atlas by integrating public datasets and in-house spatial omics data generated for human tissues. To analyse spatial omics for cell-cell interactions, we develop deep graph neural network models to predict which cell types are interacting via which ligand-receptor pairs.
1.Image processing and computer vision for processing the image data of spatial omics
2.Graph neural network
3.Social network analysis
Artificial Intelligent (AI)/Bioinformatician research fellow (postdoctoral) positions are available in Jinmiao Chen's (CJM) lab. Characterization of immune cell heterogeneity and cell-cell interactions at single-cell resolution is critical for unders
Skills: Research Fellow (AI/Bioinformatician) / SIgN
Experience: 0.00-50.00 Years
The number of jobs in each salary range for all: