#39 · Single-cell

Local LLMs and modular frameworks standardize single-cell metadata, model use and…

The shared thread is standardization of single-cell analysis inputs, interfaces, and annotation workflows rather than new biological inference. h5adify harmonizes metadata and gene identifiers across AnnData H5AD files by pairing deterministic biological inference with local LLMs. scUnify separates model-specific trainers, downstream tasks, and adaptation strategies so single-cell foundation models can be run zero-shot and compared. CellTypeAI applies a locally hosted generative-AI program with retrieval augmented generation for tissue-context cell type annotation of scRNA-seq data. scEvolver accumulates cell-type prototypes for continual annotation without revisiting historical data. What is new is the combination of language-model components with explicit structure, as in h5adify's deterministic grounding and CellTypeAI's retrieval step, alongside modular organization in scUnify.