The Factorized Embeddings Model: Towards a Data-driven Cell Atlas
As sequencing technologies get more precise, the definition of a cell type becomes vague and less relevant. In her work, Assya Trofimov proposes a method to compute continuous embeddings for biological variation without the need for hard-coded rules. Since cells have diverse simultaneous facets, from discrete to continuous cell types and states, approaching the construction of a cell atlas in a data-driven manner is key. Assya’s thesis project is about creating a human cell atlas that will help define better the concept of cell identity.
Keywords: data-driven cell atlas, cell atlas, representation learning, deep learning, machine learning, transcriptomics, TCR-Seq