Summary of Cell-ontology Guided Transcriptome Foundation Model, by Xinyu Yuan et al.
Cell-ontology guided transcriptome foundation model
by Xinyu Yuan, Zhihao Zhan, Zuobai Zhang, Manqi Zhou, Jianan Zhao, Boyu Han, Yue Li, Jian Tang
First submitted to arxiv on: 22 Aug 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Artificial Intelligence (cs.AI)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary The paper proposes a new approach to training transcriptome foundation models (TFMs) that leverages taxonomic relationships between cell types in cell ontology graphs. Current TFMs treat cells as independent samples, ignoring the available ontological information. The authors argue that incorporating this information during pre-training can improve learning of biologically meaningful gene co-expression patterns and enhance general-purpose foundation model performance for zero-shot and fine-tuning tasks. They introduce a novel loss component, cell-type coherence loss, which guides the TFM to learn cell-type-specific representations. The approach is evaluated on 22 million cells from the CellxGene database and demonstrates competitive generalization and transferability performance over existing TFMs on biologically important tasks such as identifying novel cell types, predicting marker genes, and cancer drug responses. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper explores a new way to train special kinds of artificial intelligence models that help us understand human diseases. These models are trained using large amounts of data from individual cells in the body. The current approach treats each cell as separate, but this paper suggests that by considering relationships between different cell types, we can make these models more accurate and useful for understanding complex biological processes. The researchers developed a new technique to train these models, which they tested on a huge dataset of cell information. Their results show that their approach is effective in identifying specific cells and predicting how certain diseases will respond to treatments. |
Keywords
» Artificial intelligence » Fine tuning » Generalization » Transferability » Zero shot
