Top AI Repos — open-source AI, indexed and scored
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
State is a machine learning model that predicts cellular perturbation response across diverse contexts
| Date | Stars |
|---|---|
| 2026-07-31 | 624 |
| 2026-08-02 | 627 |
| 2026-08-06 | 627 |
Today
— stars today
This week
— stars this week
This month
— stars this month
Momentum
15.0
growth rate 0.00%/day
# Predicting cellular responses to perturbation across diverse contexts with State > Train State transition models or pretrain State embedding models. See the State [paper](https://www.biorxiv.org/content/10.1101/2025.06.26.661135v2). > > See the [Google Colab](https://colab.research.google.com/drive/1QKOtYP7bMpdgDJEipDxaJqOchv7oQ-_l) to train STATE for the [Virtual Cell Challenge](https://virtualcellchallenge.org/). ## Associated repositories - Model evaluation framework: [cell-eval](https://github.com/ArcInstitute/cell-eval) - Dataloaders and preprocessing: [cell-load](https://github.com/ArcInstitute/cell-load) ## Getting started - Train an ST model for genetic perturbation prediction using the Replogle-Nadig dataset: [Colab](https://colab.research.google.com/drive/1Ih-KtTEsPqDQnjTh6etVv_f-gRAA86ZN) - Perform inference using an ST model trained on Tahoe-100M: [Colab](https://colab.research.google.com/drive/1bq5v7hixnM-tZHwNdgPiuuDo6kuiwLKJ) - Embed and annotate a new dataset using SE: [Colab](https://colab.research.google.com/drive/1uJinTJLSesJeot0mP254fQpSxGuDEsZt) - Train STATE for the Virtual Cell Challenge: [Colab](https://colab.research.google.com/drive/1QKOtYP7bMpdgDJEipDxaJqOchv7oQ-_l) ## Installation ### Installation from PyPI This package is distributed via [`uv`](https://docs.astral.sh/uv). ```bash uv tool install arc-state ``` ### Installation from Source ```bash git clone [email protected]:ArcInstitute/state.git cd state uv run state ``` When making fundamental changes to State, install an editable version with the `-e` flag. ```bash git clone [email protected]:ArcInstitute/state.git cd state uv tool install -e . ``` ## CLI Usage If installed via `uv tool install`, run `state ...`. From source, run `uv run state ...`. Use `state --help` (or `state tx --help`, `state emb --help`) to see available subcommands. ## State Transition Model (ST) Use `state tx` to train and run perturbation prediction models. ### preprocess_train Prepares training h5ad files by normalizing counts, applying log1p, and selecting highly variable genes (HVGs). The HVG matrix is stored in `.obsm["X_hvg"]`, and the output .h5ad is written to `--output`. ```bash state tx preprocess_train \ --adata /path/to/raw_data.h5ad \ --output /path/to/preprocessed_training_data.h5ad \ --num_hvgs 2000 ``` ### train Trains an ST model using Hydra overrides. Point `data.kwargs.toml_config_path` at a TOML file that defines datasets and splits. `output_dir` and `name` define the run directory (`output_dir/name`). ```bash state tx train \ data.kwargs.toml_config_path=examples/fewshot.toml \ data.kwargs.embed_key=X_hvg \ data.kwargs.pert_col=target_gene \ data.kwargs.cell_type_key=cell_type \ training.max_steps=40000 \ training.batch_size=8 \ model=state \ output_dir="$HOME/state" \ name="test" ``` Optional downsampling overrides: - `data.kwargs.downsample` downsample counts during loading (`<=1` keeps that fraction by binomial sampling, `>1` targets that read depth per cell for `output_space=all`). - `data.kwargs.downsample_cells` caps the number of cells loaded per `(cell_type, perturbation[, batch])` group while leaving smaller groups unchanged. ### predict Evaluates a trained run with `cell-eval` metrics (or just runs prediction with `--predict-only`). `--output-dir` should point to the run directory that contains `config.yaml` and `checkpoints/`. ```bash state tx predict --output-dir $HOME/state/test --checkpoint final.ckpt ``` Use `--toml` to evaluate on a different dataset/split config than the one saved in the run. ### infer Runs inference on new data (not necessarily in the training TOML). Provide the run directory via `--model-dir` and an input AnnData via `--adata`. Use `--embed-key X_hvg` if you trained on HVG features. ```bash state tx infer \ --model-dir /path/to/run \ --checkpoint /path/to/run/checkpoints/final.ckpt \ --adata /path/to/preprocessed_data.h5ad \ --pert-col gene \ --embed-key
Excerpt of 9,208 characters
Read on GitHub
446
113
81
42
39
29
25
18
15
6
5
5
Christopher Fleetwood · United Kingdom
2
2
1
1
Would you bet a product on this? Bounded 0–100 and slow moving.
matched fp:8bfd70557f3aabca, llm:Repository description: 'State is a machine learning model that predicts cellular perturbation response across diverse contexts' (Python).
matched fp:8bfd70557f3aabca, llm:Repository description: 'State is a machine learning model that predicts cellular perturbation response across diverse contexts' (Python).