deepdraw-opt 0.1.1__tar.gz

deepdraw_opt-0.1.1/.gitignore

@@ -0,0 +1,135 @@
+# Python
+__pycache__/
+*.py[cod]
+*$py.class
+*.so
+.Python
+build/
+develop-eggs/
+dist/
+downloads/
+eggs/
+.eggs/
+lib/
+lib64/
+parts/
+sdist/
+var/
+wheels/
+*.egg-info/
+.installed.cfg
+*.egg
+
+# Virtual Environment
+venv/
+env/
+ENV/
+.env
+.venv
+.venv*/
+env.bak/
+venv.bak/
+
+# IDE
+.idea/
+.vscode/
+*.swp
+*.swo
+.DS_Store
+
+# Jupyter Notebook
+.ipynb_checkpoints
+*.ipynb
+
+# Testing
+.coverage
+htmlcov/
+.pytest_cache/
+.tox/
+
+# Distribution
+*.tar.gz
+*.zip
+
+# los and databases
+*.log
+*.sqlite
+*.db
+*.csv
+!examples/deepdraw_dummy/design_pool.csv
+!examples/deepdraw_dummy/measurements.csv
+*.pdf
+*.png
+*.err
+*.out
+
+# Model files and data
+*.h5
+*.pkl
+*.model
+*.pth
+*.7z
+*.safetensors
+multirun/*
+data/*
+data_new/*
+!data/scripts/*
+models/
+checkpoints/
+results/
+results_*/
+runs/
+deepdraw_run/
+wandb/
+data_assembly/
+data_analyze/
+
+# Environment management
+.python-version
+poetry.lock
+.ruff_cache/
+
+# Documentation
+docs/_build/
+site/
+
+# Misc
+.DS_Store
+.env.local
+.env.development.local
+.env.test.local
+.env.production.local
+# run_experiments.py
+setting.json
+
+# dir
+data_new_deprecated/
+
+# logs
+logs
+multirun
+outputs
+logs_experiments
+log_test/*
+logs_test/*
+slurm_logs/
+
+
+# hidden directories
+.*
+
+# Local assistant settings and instructions
+.codex/
+.claude/
+AGENTS.md
+CLAUDE.md
+
+# local data
+**/test_local.yaml
+**/local_166k.yaml
+/benchmarks/datasets/
+
+# Local-only analysis / figure-generation scripts (notebooks, sensitivity-analysis
+# helpers, etc.). Anything dropped under this folder stays out of git.
+benchmarks/utils/analysis/
+/utils/

deepdraw_opt-0.1.1/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 Cell Ethology Lab and contributors
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

deepdraw_opt-0.1.1/PKG-INFO

@@ -0,0 +1,268 @@
+Metadata-Version: 2.5
+Name: deepdraw-opt
+Version: 0.1.1
+Summary: Active learning for genetic circuit design
+Project-URL: Homepage, https://github.com/cellethology/deepdraw
+Project-URL: Repository, https://github.com/cellethology/deepdraw.git
+Project-URL: Issues, https://github.com/cellethology/deepdraw/issues
+Author-email: Cell Ethology Lab <jerry@westlake.edu.cn>
+License: MIT
+License-File: LICENSE
+Keywords: active-learning,circuit-design,dna-sequences,machine-learning
+Classifier: Development Status :: 3 - Alpha
+Classifier: Intended Audience :: Science/Research
+Classifier: License :: OSI Approved :: MIT License
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3.10
+Classifier: Programming Language :: Python :: 3.11
+Classifier: Topic :: Scientific/Engineering :: Bio-Informatics
+Requires-Python: >=3.10
+Requires-Dist: botorch>=0.16.1
+Requires-Dist: gpytorch>=1.14.2
+Requires-Dist: hydra-core>=1.3.2
+Requires-Dist: numpy>=1.21.0
+Requires-Dist: omegaconf>=2.3.0
+Requires-Dist: pandas>=1.3.0
+Requires-Dist: scikit-learn<1.8,>=1.6.1
+Requires-Dist: scipy>=1.6.0
+Requires-Dist: torch>=2.0.1
+Provides-Extra: analysis
+Requires-Dist: openpyxl>=3.1.5; extra == 'analysis'
+Requires-Dist: pyyaml>=6.0.2; extra == 'analysis'
+Requires-Dist: tqdm<=4.66.5; extra == 'analysis'
+Provides-Extra: cluster
+Requires-Dist: hydra-submitit-launcher>=1.2.0; extra == 'cluster'
+Requires-Dist: submitit>=1.5.3; extra == 'cluster'
+Description-Content-Type: text/markdown
+
+# Deepdraw
+
+[![CI](https://github.com/cellethology/deepdraw/actions/workflows/pre-commit.yml/badge.svg)](https://github.com/cellethology/deepdraw/actions/workflows/pre-commit.yml)
+[![Coverage](https://img.shields.io/codecov/c/github/cellethology/deepdraw?logo=codecov)](https://codecov.io/gh/cellethology/deepdraw)
+
+Deepdraw selects batches of genetic circuits to measure using sequence embeddings and accumulated experimental results. Its default search combines ProbCover initial selection, Gaussian-process prediction, and maximum-value entropy search (MES).
+
+The active-learning loop runs on CPU. AlphaGenome embedding generation has a separate optional environment. For retrospective evaluation on fully measured datasets, see [benchmarks](https://github.com/cellethology/deepdraw/blob/main/benchmarks/README.md).
+
+## Install
+
+Install with Python 3.10 or later:
+
+```bash
+python -m pip install deepdraw-opt
+deepdraw --help
+```
+
+The package name is `deepdraw-opt`; the command and Python import are `deepdraw`.
+
+For the bundled example, optional AlphaGenome environment, and development, install Git and [uv](https://docs.astral.sh/uv/), then clone the repository:
+
+```bash
+git clone https://github.com/cellethology/deepdraw.git
+cd deepdraw
+uv sync --python 3.10
+uv run deepdraw --help
+```
+
+The commands below use `uv run` from the repository root. With a PyPI installation, invoke `deepdraw` directly instead of `uv run deepdraw`.
+
+## Try the dummy example
+
+The example includes 60 designs, synthetic embeddings, and measurements matching the default first batch. It requires no model download.
+
+```bash
+uv run deepdraw init \
+  --pool-csv examples/deepdraw_dummy/design_pool.csv \
+  --embeddings examples/deepdraw_dummy/embeddings.npz \
+  --sequence-column sequence \
+  --id-column variant_id \
+  --stopping-rule threshold \
+  --phenotype-threshold 19
+
+uv run deepdraw suggest \
+  --run-dir deepdraw_run \
+  --measurements examples/deepdraw_dummy/measurements.csv \
+  --label-column Expression
+```
+
+These commands write `deepdraw_run/round_000_to_measure.csv` and `round_001_to_measure.csv`. See the [example README](https://github.com/cellethology/deepdraw/blob/main/examples/deepdraw_dummy/README.md) to simulate another measurement update and exercise stopping.
+
+## Use your own data
+
+### 1. Prepare the design pool and embeddings
+
+Create `designs.csv` with one row per candidate, a unique ID, and its DNA sequence:
+
+```csv
+variant_id,sequence
+variant_001,ATGCGTACGTTAGCGA
+variant_002,ATGCGTACGATAGCAA
+variant_003,ATGCGTACGCTAGCTA
+```
+
+Provide an NPZ containing:
+
+| Array | Contents |
+| --- | --- |
+| `embeddings` | A numeric array of shape `(number_of_designs, number_of_features)`. |
+| `ids` | Design IDs matching the CSV's ID column, one per embedding row. |
+
+`sample_ids` is also accepted as the ID-array name. Zero-based pool row indices are supported; an NPZ without IDs must follow the CSV's row order exactly. Keep the pool and embeddings at their original paths while continuing a run.
+
+You can use embeddings from another model or generate them with the [AlphaGenome workflow below](#generate-alphagenome-embeddings).
+
+### 2. Select the first batch
+
+```bash
+uv run deepdraw init \
+  --pool-csv designs.csv \
+  --embeddings embeddings.npz \
+  --sequence-column sequence \
+  --id-column variant_id \
+  --output-dir runs/my_project \
+  --stopping-rule threshold \
+  --phenotype-threshold 10
+```
+
+Choose a phenotype cutoff appropriate for your experiment. Deepdraw maximizes the supplied score.
+
+The run directory contains:
+
+```text
+runs/my_project/
+├── deepdraw_state.json
+├── round_000_to_measure.csv
+├── latest_recommendations.csv
+└── selection_history.csv
+```
+
+Measure the designs in `round_000_to_measure.csv`. Recommendation files retain the original pool columns; `selection_history.csv` also records `deepdraw_round`.
+
+### 3. Add measurements and request the next batch
+
+Create a cumulative `measurements.csv` containing design IDs and one measured score per design:
+
+```csv
+variant_id,Expression
+variant_001,1.42
+variant_002,3.87
+variant_003,5.11
+```
+
+Use the IDs actually recommended for your run. Include every previously recommended design; additional measured designs from the pool can also be included.
+
+```bash
+uv run deepdraw suggest \
+  --run-dir runs/my_project \
+  --measurements measurements.csv \
+  --label-column Expression
+```
+
+Deepdraw trains on the measured designs and writes `round_001_to_measure.csv`. Measure that batch, append its scores to the same cumulative file, and repeat `suggest` for later rounds. Measured designs are excluded from new batches.
+
+When a stopping rule fires, the command reports the reason and writes no new batch. Use a new `--output-dir` when starting another run.
+
+## Search settings and stopping
+
+The CLI selects the recommended search settings automatically. Algorithm flags are optional overrides:
+
+| Flag | Default | Alternatives |
+| --- | --- | --- |
+| `--starting-batch-size`, `--batch-size` | `12` | Any positive integer; capped by available designs. |
+| `--seed` | `0` | An integer from `0` to `4294967295`. |
+| `--initial-selection-strategy` | `probcover` | `core_set`, `density_core_set`, `kmeans`, `probcover_cosine`, `typiclust`, `random`. |
+| `--predictor` | `gp` | `linear_gp`, `gradboost_regressor`, `rf_ensemble`. |
+| `--query-strategy` | `mes` | `topk`, `predstdhybrid`, `ucb`, `pes`, `gibbon`, `log_pi`, `log_ei`, `qlog_nei`, `qlog_ei`, `random`. |
+| `--feature-transforms` | `standardize` | `none`. |
+| `--target-transforms` | `log_standardize` | `none`. |
+| `--stopping-rule` | `none` | `threshold`, `ei`. |
+
+`probcover` uses Euclidean distance; `probcover_cosine` uses cosine distance.
+
+Gradient boosting supports `topk`, `predstdhybrid`, and `random`. The random-forest ensemble also supports `ucb`, `log_pi`, `log_ei`, `qlog_nei`, and `qlog_ei`. The entropy queries `mes`, `pes`, and `gibbon` require `gp` or `linear_gp`.
+
+The default target transform applies `log(1 + score)` followed by standardization. Use `--target-transforms none` for scores that cannot be log-transformed.
+
+Fixed-threshold stopping, used above, stops when a measured score meets or exceeds `--phenotype-threshold`. For an adaptive EI example using the bundled data:
+
+```bash
+uv run deepdraw init \
+  --pool-csv examples/deepdraw_dummy/design_pool.csv \
+  --embeddings examples/deepdraw_dummy/embeddings.npz \
+  --sequence-column sequence \
+  --id-column variant_id \
+  --output-dir runs/dummy_ei \
+  --stopping-rule ei \
+  --epsilon 0.01
+
+uv run deepdraw suggest \
+  --run-dir runs/dummy_ei \
+  --measurements examples/deepdraw_dummy/measurements.csv \
+  --label-column Expression
+```
+
+After each measured batch, `suggest` stops if maximum expected improvement over the unmeasured pool is at most epsilon. With the default target transform, epsilon is in `log(1 + score)` units before standardization. The value above is for the dummy example; choose it for your objective. EI stopping reuses a compatible fitted GP or fits a separate GP when another predictor is selected.
+
+Omitting the stopping options disables threshold and EI stopping. All runs finish when no unmeasured designs remain. Use `--log-level WARNING` for quieter output.
+
+## Generate AlphaGenome embeddings
+
+Skip this section if you already have embeddings. The isolated environment uses Python 3.11:
+
+```bash
+uv sync --project envs/alphagenome --python 3.11
+```
+
+Accept access to `google/alphagenome-all-folds` on Hugging Face, then authenticate:
+
+```bash
+uv run --project envs/alphagenome hf auth login
+```
+
+An `HF_TOKEN` environment variable is also supported. The embedding environment supports CUDA GPUs on Linux and CPU execution; use CPU on macOS. Start with a batch size of 1.
+
+```bash
+uv run --project envs/alphagenome deepdraw embed-alphagenome \
+  --pool-csv designs.csv \
+  --sequence-column sequence \
+  --id-column variant_id \
+  --output embeddings_alphagenome_raw.npz \
+  --model-version all_folds \
+  --resolution 1 \
+  --species human \
+  --pooling mean \
+  --batch-size 1
+
+uv run deepdraw pca \
+  --input embeddings_alphagenome_raw.npz \
+  --output embeddings.npz \
+  --selection-method kneedle
+```
+
+Use the resulting `embeddings.npz` with `deepdraw init`. PCA runs in the main Deepdraw environment.
+
+## Repository layout and development
+
+| Directory | Contents |
+| --- | --- |
+| [deepdraw/](https://github.com/cellethology/deepdraw/tree/main/deepdraw) | CLI, experimental workflow, embedding tools, and shared presets. |
+| [core/](https://github.com/cellethology/deepdraw/blob/main/core/README.md) | Models, selection strategies, training, and stopping rules. |
+| [examples/deepdraw_dummy/](https://github.com/cellethology/deepdraw/blob/main/examples/deepdraw_dummy/README.md) | Runnable example with synthetic data. |
+| [benchmarks/](https://github.com/cellethology/deepdraw/blob/main/benchmarks/README.md) | Retrospective evaluation, noise simulation, and Hydra/Slurm tools. |
+| [envs/alphagenome/](https://github.com/cellethology/deepdraw/tree/main/envs/alphagenome) | Optional AlphaGenome environment. |
+| [test/](https://github.com/cellethology/deepdraw/tree/main/test) | Unit and workflow tests. |
+
+Install development dependencies and run the tests:
+
+```bash
+uv run --group dev pytest
+```
+
+Command-specific help:
+
+```bash
+uv run deepdraw init --help
+uv run deepdraw suggest --help
+uv run deepdraw pca --help
+uv run --project envs/alphagenome deepdraw embed-alphagenome --help
+```

deepdraw_opt-0.1.1/README.md

@@ -0,0 +1,231 @@
+# Deepdraw
+
+[![CI](https://github.com/cellethology/deepdraw/actions/workflows/pre-commit.yml/badge.svg)](https://github.com/cellethology/deepdraw/actions/workflows/pre-commit.yml)
+[![Coverage](https://img.shields.io/codecov/c/github/cellethology/deepdraw?logo=codecov)](https://codecov.io/gh/cellethology/deepdraw)
+
+Deepdraw selects batches of genetic circuits to measure using sequence embeddings and accumulated experimental results. Its default search combines ProbCover initial selection, Gaussian-process prediction, and maximum-value entropy search (MES).
+
+The active-learning loop runs on CPU. AlphaGenome embedding generation has a separate optional environment. For retrospective evaluation on fully measured datasets, see [benchmarks](https://github.com/cellethology/deepdraw/blob/main/benchmarks/README.md).
+
+## Install
+
+Install with Python 3.10 or later:
+
+```bash
+python -m pip install deepdraw-opt
+deepdraw --help
+```
+
+The package name is `deepdraw-opt`; the command and Python import are `deepdraw`.
+
+For the bundled example, optional AlphaGenome environment, and development, install Git and [uv](https://docs.astral.sh/uv/), then clone the repository:
+
+```bash
+git clone https://github.com/cellethology/deepdraw.git
+cd deepdraw
+uv sync --python 3.10
+uv run deepdraw --help
+```
+
+The commands below use `uv run` from the repository root. With a PyPI installation, invoke `deepdraw` directly instead of `uv run deepdraw`.
+
+## Try the dummy example
+
+The example includes 60 designs, synthetic embeddings, and measurements matching the default first batch. It requires no model download.
+
+```bash
+uv run deepdraw init \
+  --pool-csv examples/deepdraw_dummy/design_pool.csv \
+  --embeddings examples/deepdraw_dummy/embeddings.npz \
+  --sequence-column sequence \
+  --id-column variant_id \
+  --stopping-rule threshold \
+  --phenotype-threshold 19
+
+uv run deepdraw suggest \
+  --run-dir deepdraw_run \
+  --measurements examples/deepdraw_dummy/measurements.csv \
+  --label-column Expression
+```
+
+These commands write `deepdraw_run/round_000_to_measure.csv` and `round_001_to_measure.csv`. See the [example README](https://github.com/cellethology/deepdraw/blob/main/examples/deepdraw_dummy/README.md) to simulate another measurement update and exercise stopping.
+
+## Use your own data
+
+### 1. Prepare the design pool and embeddings
+
+Create `designs.csv` with one row per candidate, a unique ID, and its DNA sequence:
+
+```csv
+variant_id,sequence
+variant_001,ATGCGTACGTTAGCGA
+variant_002,ATGCGTACGATAGCAA
+variant_003,ATGCGTACGCTAGCTA
+```
+
+Provide an NPZ containing:
+
+| Array | Contents |
+| --- | --- |
+| `embeddings` | A numeric array of shape `(number_of_designs, number_of_features)`. |
+| `ids` | Design IDs matching the CSV's ID column, one per embedding row. |
+
+`sample_ids` is also accepted as the ID-array name. Zero-based pool row indices are supported; an NPZ without IDs must follow the CSV's row order exactly. Keep the pool and embeddings at their original paths while continuing a run.
+
+You can use embeddings from another model or generate them with the [AlphaGenome workflow below](#generate-alphagenome-embeddings).
+
+### 2. Select the first batch
+
+```bash
+uv run deepdraw init \
+  --pool-csv designs.csv \
+  --embeddings embeddings.npz \
+  --sequence-column sequence \
+  --id-column variant_id \
+  --output-dir runs/my_project \
+  --stopping-rule threshold \
+  --phenotype-threshold 10
+```
+
+Choose a phenotype cutoff appropriate for your experiment. Deepdraw maximizes the supplied score.
+
+The run directory contains:
+
+```text
+runs/my_project/
+├── deepdraw_state.json
+├── round_000_to_measure.csv
+├── latest_recommendations.csv
+└── selection_history.csv
+```
+
+Measure the designs in `round_000_to_measure.csv`. Recommendation files retain the original pool columns; `selection_history.csv` also records `deepdraw_round`.
+
+### 3. Add measurements and request the next batch
+
+Create a cumulative `measurements.csv` containing design IDs and one measured score per design:
+
+```csv
+variant_id,Expression
+variant_001,1.42
+variant_002,3.87
+variant_003,5.11
+```
+
+Use the IDs actually recommended for your run. Include every previously recommended design; additional measured designs from the pool can also be included.
+
+```bash
+uv run deepdraw suggest \
+  --run-dir runs/my_project \
+  --measurements measurements.csv \
+  --label-column Expression
+```
+
+Deepdraw trains on the measured designs and writes `round_001_to_measure.csv`. Measure that batch, append its scores to the same cumulative file, and repeat `suggest` for later rounds. Measured designs are excluded from new batches.
+
+When a stopping rule fires, the command reports the reason and writes no new batch. Use a new `--output-dir` when starting another run.
+
+## Search settings and stopping
+
+The CLI selects the recommended search settings automatically. Algorithm flags are optional overrides:
+
+| Flag | Default | Alternatives |
+| --- | --- | --- |
+| `--starting-batch-size`, `--batch-size` | `12` | Any positive integer; capped by available designs. |
+| `--seed` | `0` | An integer from `0` to `4294967295`. |
+| `--initial-selection-strategy` | `probcover` | `core_set`, `density_core_set`, `kmeans`, `probcover_cosine`, `typiclust`, `random`. |
+| `--predictor` | `gp` | `linear_gp`, `gradboost_regressor`, `rf_ensemble`. |
+| `--query-strategy` | `mes` | `topk`, `predstdhybrid`, `ucb`, `pes`, `gibbon`, `log_pi`, `log_ei`, `qlog_nei`, `qlog_ei`, `random`. |
+| `--feature-transforms` | `standardize` | `none`. |
+| `--target-transforms` | `log_standardize` | `none`. |
+| `--stopping-rule` | `none` | `threshold`, `ei`. |
+
+`probcover` uses Euclidean distance; `probcover_cosine` uses cosine distance.
+
+Gradient boosting supports `topk`, `predstdhybrid`, and `random`. The random-forest ensemble also supports `ucb`, `log_pi`, `log_ei`, `qlog_nei`, and `qlog_ei`. The entropy queries `mes`, `pes`, and `gibbon` require `gp` or `linear_gp`.
+
+The default target transform applies `log(1 + score)` followed by standardization. Use `--target-transforms none` for scores that cannot be log-transformed.
+
+Fixed-threshold stopping, used above, stops when a measured score meets or exceeds `--phenotype-threshold`. For an adaptive EI example using the bundled data:
+
+```bash
+uv run deepdraw init \
+  --pool-csv examples/deepdraw_dummy/design_pool.csv \
+  --embeddings examples/deepdraw_dummy/embeddings.npz \
+  --sequence-column sequence \
+  --id-column variant_id \
+  --output-dir runs/dummy_ei \
+  --stopping-rule ei \
+  --epsilon 0.01
+
+uv run deepdraw suggest \
+  --run-dir runs/dummy_ei \
+  --measurements examples/deepdraw_dummy/measurements.csv \
+  --label-column Expression
+```
+
+After each measured batch, `suggest` stops if maximum expected improvement over the unmeasured pool is at most epsilon. With the default target transform, epsilon is in `log(1 + score)` units before standardization. The value above is for the dummy example; choose it for your objective. EI stopping reuses a compatible fitted GP or fits a separate GP when another predictor is selected.
+
+Omitting the stopping options disables threshold and EI stopping. All runs finish when no unmeasured designs remain. Use `--log-level WARNING` for quieter output.
+
+## Generate AlphaGenome embeddings
+
+Skip this section if you already have embeddings. The isolated environment uses Python 3.11:
+
+```bash
+uv sync --project envs/alphagenome --python 3.11
+```
+
+Accept access to `google/alphagenome-all-folds` on Hugging Face, then authenticate:
+
+```bash
+uv run --project envs/alphagenome hf auth login
+```
+
+An `HF_TOKEN` environment variable is also supported. The embedding environment supports CUDA GPUs on Linux and CPU execution; use CPU on macOS. Start with a batch size of 1.
+
+```bash
+uv run --project envs/alphagenome deepdraw embed-alphagenome \
+  --pool-csv designs.csv \
+  --sequence-column sequence \
+  --id-column variant_id \
+  --output embeddings_alphagenome_raw.npz \
+  --model-version all_folds \
+  --resolution 1 \
+  --species human \
+  --pooling mean \
+  --batch-size 1
+
+uv run deepdraw pca \
+  --input embeddings_alphagenome_raw.npz \
+  --output embeddings.npz \
+  --selection-method kneedle
+```
+
+Use the resulting `embeddings.npz` with `deepdraw init`. PCA runs in the main Deepdraw environment.
+
+## Repository layout and development
+
+| Directory | Contents |
+| --- | --- |
+| [deepdraw/](https://github.com/cellethology/deepdraw/tree/main/deepdraw) | CLI, experimental workflow, embedding tools, and shared presets. |
+| [core/](https://github.com/cellethology/deepdraw/blob/main/core/README.md) | Models, selection strategies, training, and stopping rules. |
+| [examples/deepdraw_dummy/](https://github.com/cellethology/deepdraw/blob/main/examples/deepdraw_dummy/README.md) | Runnable example with synthetic data. |
+| [benchmarks/](https://github.com/cellethology/deepdraw/blob/main/benchmarks/README.md) | Retrospective evaluation, noise simulation, and Hydra/Slurm tools. |
+| [envs/alphagenome/](https://github.com/cellethology/deepdraw/tree/main/envs/alphagenome) | Optional AlphaGenome environment. |
+| [test/](https://github.com/cellethology/deepdraw/tree/main/test) | Unit and workflow tests. |
+
+Install development dependencies and run the tests:
+
+```bash
+uv run --group dev pytest
+```
+
+Command-specific help:
+
+```bash
+uv run deepdraw init --help
+uv run deepdraw suggest --help
+uv run deepdraw pca --help
+uv run --project envs/alphagenome deepdraw embed-alphagenome --help
+```