부산대학교 도서관

Integrating multimodal single-cell data is formalized into predicting co-variation among DNA, RNA, and protein measurements within individual cells. Accurate prediction of single-cell table data across multiple modalities will provide a more in-depth understanding of the mechanisms underlying tissue function or dysfunction in health and disease. While previous studies highlight the potential of machine learning in multimodal single-cell integration, there is a lack of comparison studies that focus on technical aspects rather than biology-related details. Therefore, this study encompassed the entire pipeline and compared the four highest-performing methods in the Multimodal single-cell data integration NeurIPS 2022 Competition through performance and efficiency, which provides data scientists with appropriate entry points. Experiment results revealed the significance of preprocessing and feature engineering in addressing problems and consolidating the trend of using machine learning techniques such as the Variational Autoencoder (VAE) and ensemble model.