DOGMA: Weaving Structural Information into Data-centric Single-cell Transcriptomics Analysis
Researchers have introduced DOGMA, a novel data-centric AI framework for single-cell transcriptomics analysis. This framework integrates multi-level biological prior knowledge, moving beyond purely data-driven heuristics. DOGMA constructs biologically grounded cell graphs by incorporating statistical alignment with Cell Ontology and phylogenetic structure, and enhances feature-level semantics using Gene Ontology. The system demonstrates robustness in zero-shot cell-type evaluation and sample efficiency across complex multi-species benchmarks, while requiring less GPU memory and inference time. AI
IMPACT Provides a new method for enhancing biological data analysis with AI, potentially improving efficiency and accuracy in transcriptomics research.