Deep learning-based data imputation and missing modality prediction for single-cell multi-omics data

Recent advancements in single-cell multimodal profiling technologies enable various molecular programs such as gene expression, protein abundance, and chromatin accessibility to be profiled in individual cells. While the single-cell multimodal omics data generated from these technologies hold promise to model and understand cell types and cell fate decisions at an unprecedented resolution, such data often contain large amounts of missing values and missing modalities.

A promising application of machine learning methods, in particular deep learning models, is to impute missing values and data modalities for such data so as to significantly improve the utility of single-cell multimodal omics data for understanding complex biological and cellular systems. This project aims to design, evaluate, and benchmark machine/deep learning methods for data imputation and for predicting missing modalities in single-cell multimodal omics data.