アブストラクト
| Date | 14:00-15:00 Jul 31, 2024 |
| Speaker | Han Zhang Chair of the Department of Intelligence Engineering College of Artificial Intelligence, Nankai University, China |
| Title | Deep Learning Model for Single-cell Proteomics Embedding and Multi-omics integration: scPROTEIN, scMHNN and BERMAD |
| Abstract |
The unique complexity of single-cell proteomic data makes the
analysis and processing of proteomic data a serious challenge. ScPROTEIN
is a deep learning framework designed specifically for single-cell
proteomic data analysis, aimed at solving a series of complex problems
in proteomic data, such as uncertainty evaluation, data-missing, batch
effects, and high noise, which are tangled during processing. Based on
deep learning, we previously designed batch effect removal model (HDMC)
and Graph pooling model (DMIPool). Similarly starting from deep
learning, scPROTEIN utilizes graph contrastive learning for single-cell
proteomics representation learning, and designs a unified deep learning
framework to learn accurate cell embedding used for a range of
downstream analyses. This framework can improve the accuracy of
single-cell proteomic data analysis, also promote the application
research of clinical proteomic data, providing new methods and tools for
analyzing complex biological problems. In addition, single-cell
multi-omics integration has become an important requirement, and the
hyper-graph neural network model scMHNN has been designed for integrated
analysis of single-cell epigenomic, transcriptomic and proteomic data.
Also, the batch effect removal model of multi-layer adaptive automatic
encoder with dual channel framework BERMAD can remove the batch effect
of single cell RNA-seq data and maintain the diversity of data. The
above research provides some examples of deep learning to solve
relatively complex problems in bioinformatics.
|