SMACT 4.0.0__tar.gz → 4.0.2__tar.gz

{smact-4.0.0 → smact-4.0.2}/LICENSE

@@ -19,3 +19,6 @@ 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.
+
+This project bundles third-party material under its own license terms;
+see THIRD_PARTY_NOTICES.md.

{smact-4.0.0/SMACT.egg-info → smact-4.0.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: SMACT
-Version: 4.0.0
+Version: 4.0.2
 Summary: Semiconducting Materials by Analogy and Chemical Theory
 Author-email: The SMACT Developers <a.walsh@imperial.ac.uk>
 Maintainer-email: "Kinga O. Mastej" <k.mastej24@imperial.ac.uk>
@@ -25,7 +25,7 @@ Description-Content-Type: text/markdown
 License-File: LICENSE
 Requires-Dist: ase
 Requires-Dist: numpy<3
-Requires-Dist: pandarallel[optional]>=1.6.5
+Requires-Dist: pandarallel>=1.6.5
 Requires-Dist: pandas
 Requires-Dist: pathos
 Requires-Dist: pymatgen>=2024.2.20
@@ -70,7 +70,7 @@ Requires-Dist: smact[featurisers]; extra == "ml"
 Provides-Extra: property-prediction
 Requires-Dist: aviary-models>=1.2.1; extra == "property-prediction"
 Requires-Dist: requests; extra == "property-prediction"
-Requires-Dist: torch>=2.0.0; extra == "property-prediction"
+Requires-Dist: torch<2.11,>=2.0.0; extra == "property-prediction"
 Provides-Extra: optional
 Requires-Dist: pydantic>=2.9.2; extra == "optional"
 Requires-Dist: smact[mp]; extra == "optional"
@@ -103,8 +103,8 @@ Requires-Dist: tqdm==4.67.1; extra == "strict"
 Dynamic: license-file
 
 [![DOI](https://joss.theoj.org/papers/10.21105/joss.01361/status.svg)](https://doi.org/10.21105/joss.01361)
-[![DOI](https://zenodo.org/badge/DOI/10.5281/zenodo.5553201.svg)](https://doi.org/10.5281/zenodo.5553201)
-[![Documentation Status](https://readthedocs.org/projects/smact/badge/?version=latest)](http://smact.readthedocs.org/en/latest/?badge=latest)
+[![DOI](https://zenodo.org/badge/DOI/10.5281/zenodo.595853.svg)](https://doi.org/10.5281/zenodo.595853)
+[![Documentation Status](https://readthedocs.org/projects/smact/badge/?version=latest)](https://smact.readthedocs.org/en/latest/?badge=latest)
 [![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://opensource.org/licenses/MIT)
 ![python version](https://img.shields.io/pypi/pyversions/smact)
 [![Ruff](https://img.shields.io/endpoint?url=https://raw.githubusercontent.com/astral-sh/ruff/main/assets/badge/v2.json)](https://github.com/astral-sh/ruff)
@@ -118,7 +118,7 @@ Dynamic: license-file
 
 # SMACT
 
-**Semiconducting Materials from Analogy and Chemical Theory** (SMACT) is a collection of rapid screening and informatics tools that uses data about chemical elements.
+**Semiconducting Materials by Analogy and Chemical Theory** (SMACT) is a collection of rapid screening and informatics tools that uses data about chemical elements.
 
 - **Documentation:** <https://smact.readthedocs.io/en/latest/>
 - **Examples:** <https://smact.readthedocs.io/en/latest/examples.html>
@@ -155,6 +155,12 @@ Use cases are available in our [examples](https://smact.readthedocs.io/en/latest
 
 - Compositions can be converted for use in Pymatgen or for representation to machine learning algorithms ([see this example](https://smact.readthedocs.io/en/latest/tutorials/smact_generation_of_solar_oxides.html)) and the related [ElementEmbeddings](https://github.com/WMD-group/ElementEmbeddings) package.
 
+- Charge neutrality screening supports **mixed-valence compounds** via the `mixed_valence=True` flag in `smact_validity`, enabling correct handling of materials like Fe₃O₄ and Mn₃O₄.
+
+- Oxidation state data is sourced from **ICSD 2024**, providing an updated and stricter set of experimentally observed oxidation states per element.
+
+- The [property prediction module](https://smact.readthedocs.io/en/latest/smact.property_prediction.html) enables composition-to-property prediction using pretrained deep learning models, including a ROOST-based band gap predictor trained on the Materials Project database.
+
 - The code also has tools for manipulating common crystal lattice types:
   - Certain structure types can be built using the [builder module](https://smact.readthedocs.io/en/latest/smact.builder.html)
   - Lattice parameters can be estimated using ionic radii of the elements for various common crystal structure types using the [lattice_parameters module](https://smact.readthedocs.io/en/latest/smact.lattice_parameters.html).
@@ -162,6 +168,88 @@ Use cases are available in our [examples](https://smact.readthedocs.io/en/latest
   - The [structure prediction](https://smact.readthedocs.io/en/latest/smact.structure_prediction.html) module can be used to predict the structure of hypothetical compositions using species similarity measures.
   - The [dopant prediction](https://smact.readthedocs.io/en/latest/smact.dopant_prediction.html) module can be used to facilitate high-throughput predictions of p-type and n-type dopants of multicomponent solids.
 
+## Package structure
+
+**Legend:** 🟢 new in v4 — 🟡 improved in v4
+
+```mermaid
+graph TD
+    classDef new fill:#c8e6c9,stroke:#388e3c,color:#000
+    classDef improved fill:#fff9c4,stroke:#f9a825,color:#000
+
+    SMACT(["smact"])
+
+    SMACT --> core["Core modules"]
+    SMACT --> SP["structure_prediction"]
+    SMACT --> DP["dopant_prediction"]
+    SMACT --> PP["🟢 property_prediction"]
+    SMACT --> IO["🟢 io"]
+    SMACT --> UT["utils"]
+
+    core --> init["init.py — Element, Species, neutral_ratios"]
+    core --> dl["data_loader.py — elemental and oxidation state data loading"]
+    core --> sc["🟡 screening.py — compositional screening"]
+    core --> pr["properties.py — band gap, electronegativity, valence electron count"]
+    core --> ox["oxidation_states.py — oxidation state combination likelihood"]
+    core --> mt["metallicity.py — metallic character scoring"]
+    core --> la["lattice.py — Site and Lattice representations"]
+    core --> bld["🟡 builder.py — perovskite and wurtzite structure builders"]
+    core --> lp["🟡 lattice_parameters.py — lattice parameter estimation from ionic radii"]
+    core --> di["distorter.py — inequivalent site enumeration and substitution"]
+
+    sc --> sc1["smact_validity — charge neutrality and Pauling electronegativity test"]
+    sc --> sc2["smact_filter — compositional search space generation"]
+    sc --> sc3["🟢 mixed_valence flag — correct handling of Fe3O4, Mn3O4"]
+    sc --> sc4["🟢 ICSD 2024 oxidation states — stricter, updated elemental data"]
+
+    bld --> bld1["cubic_perovskite — parameterized oxidation state tiling"]
+    bld --> bld2["wurtzite — corrected default cell parameters"]
+
+    lp --> lp1["corrected geometric formulae for all structure types"]
+
+    SP --> spst["structure.py — SmactStructure"]
+    SP --> spdb["database.py — StructureDB SQLite interface"]
+    SP --> spmu["mutation.py — CationMutator from lambda tables"]
+    SP --> sppd["🟡 prediction.py — StructurePredictor"]
+
+    spst --> spst1["from_file, from_mp, from_pymatgen constructors"]
+    sppd --> sppd1["ionic substitution-based crystal structure prediction"]
+    sppd --> sppd2["🟢 updated to mp_api.client.MPRester interface"]
+
+    DP --> doper["🟡 doper.py — Doper"]
+    doper --> doper1["get_dopants — p-type and n-type candidates"]
+    doper --> doper2["to_table — ranked candidates as a formatted table"]
+    doper --> doper3["plot_dopants — periodic table heatmap visualisation"]
+
+    PP --> base["base_predictor.py — BasePropertyPredictor, PredictionResult"]
+    PP --> roost["roost/ — RoostPropertyPredictor"]
+    PP --> conv["convenience.py — predict_band_gap"]
+    PP --> reg["registry.py — model discovery and resolution"]
+
+    roost --> roost1["pretrained ROOST model for band gap prediction"]
+    roost --> roost2["uncertainty estimates alongside predictions"]
+    roost --> roost3["trained on Materials Project 2024 database"]
+    base --> base1["PredictionResult — value, uncertainty, metadata"]
+
+    IO --> ee["🟢 elementembeddings.py — ElementEmbeddings interface"]
+    ee --> ee1["composition_featuriser — composition-level feature vectors"]
+    ee --> ee2["species_featuriser — species-level feature vectors"]
+
+    UT --> comp["composition.py — parse_formula, comp_maker, formula_maker"]
+    UT --> uox["🟢 oxidation.py — ICSD24OxStatesFilter"]
+    UT --> sp2["species.py — parse_spec, unparse_spec"]
+    UT --> cs["crystal_space/"]
+
+    uox --> uox1["consensus and commonality-based filtering of oxidation states"]
+    cs --> cs1["generate_composition_with_smact.py — SMACT-based composition generation"]
+    cs --> cs2["download_compounds_with_mp_api.py — Materials Project bulk download"]
+    cs --> cs3["plot_embedding.py — crystal space visualisation"]
+
+    class PP,IO new
+    class sc,bld,lp,sppd,doper,uox improved
+    class sc3,sc4,sppd2,ee,uox new
+```
+
 ## List of modules
 
 - **smact** library containing:
@@ -179,36 +267,73 @@ Use cases are available in our [examples](https://smact.readthedocs.io/en/latest
     substituting on inequivalent sites of a sub-lattice.
   - **oxidation_states.py**: Used for predicting the likelihood of species coexisting in a compound based on a statistical model.
   - **structure_prediction**: A submodule which contains a collection of tools for facilitating crystal structure predictions via ionic substitutions
-  - **dopant_prediction**: A submodule which contains a collections of tools for predicting dopants.
-  - **utils.py** A collection of utility functions used throughout the codebase.
+  - **dopant_prediction**: A submodule which contains a collection of tools for predicting dopants.
+  - **property_prediction**: A submodule for composition-to-property prediction using pretrained deep learning models (e.g. ROOST band gap predictor).
+  - **utils**: A submodule containing utility functions for composition parsing, species handling, oxidation state filtering, and crystal space generation and download.
 
 ## Requirements
 
 The main language is Python 3 and has been tested using Python 3.11 - 3.13.
-Core dependencies include NumPy, SciPy, pandas, [pymatgen](https://pymatgen.org), [ASE](https://wiki.fysik.dtu.dk/ase), and [spglib](http://atztogo.github.io/spglib). A full list is in [`pyproject.toml`](pyproject.toml).
+Core dependencies include NumPy, SciPy, pandas, [pymatgen](https://pymatgen.org), [ASE](https://wiki.fysik.dtu.dk/ase), and [spglib](https://spglib.readthedocs.io). A full list is in [`pyproject.toml`](pyproject.toml).
 
 ## Installation
 
 The latest stable release can be installed via pip:
 
-    pip install smact
+```bash
+pip install smact
+```
+
+Optional dependencies, needed by most of the examples and tutorials:
+
+```bash
+pip install "smact[optional]"
+```
 
-Optional dependencies (needed for full replication of examples and tutorials):
+The band-gap prediction example additionally needs the pre-trained ROOST models, which
+bring in PyTorch. To run every example and tutorial:
 
-    pip install "smact[optional]"
+```bash
+pip install "smact[optional,property_prediction]"
+```
 
 SMACT is also available via conda-forge:
 
-    conda install -c conda-forge smact
+```bash
+conda install -c conda-forge smact
+```
+
+The optional functionality is available as conda-forge feature packages. Install
+only the feature you need; for example, the pre-trained property-prediction
+models can be installed with:
+
+```bash
+conda install -c conda-forge smact-property-prediction
+```
+
+The other feature packages are `smact-mp`, `smact-crystal-space`,
+`smact-featurisers`, `smact-visualisation`, `smact-ml`, and `smact-optional`.
+`smact-optional` matches the `smact[optional]` pip extra. To install all
+documented examples and tutorials, including property prediction:
+
+```bash
+conda install -c conda-forge smact-optional smact-property-prediction
+```
+
+Conda feature packages use conda-forge builds of compiled dependencies such as
+`pytorch` (the package is named `torch` on PyPI). Select any GPU-specific
+`pytorch` configuration separately for your platform.
 
 ### Developer installation
 
 We use [uv](https://docs.astral.sh/uv/) for dependency management. To set up a development environment:
 
-    git clone https://github.com/wmd-group/smact.git
-    cd smact
-    uv sync --extra optional --extra property_prediction --dev
-    pre-commit install
+```bash
+git clone https://github.com/wmd-group/smact.git
+cd smact
+uv sync --extra optional --extra property_prediction --dev
+pre-commit install
+```
 
 This installs SMACT in editable mode with all optional and development dependencies, and sets up pre-commit hooks. See [CONTRIBUTING.md](CONTRIBUTING.md) for the full workflow.
 
@@ -237,14 +362,21 @@ We use [GitHub Actions](https://github.com/features/actions) for CI. Tests shoul
 
 Run the tests locally:
 
-    make test
+```bash
+make test
+```
 
 Or to run the full CI pipeline (pre-commit hooks and tests):
 
-    make ci-local
+```bash
+make ci-local
+```
 
 ## References
 
+[K. O. Mastej et al.,
+"Chemical filters for ultra-high-throughput materials screening and generation" _arXiv_ (2026)](https://doi.org/10.48550/arXiv.2607.17910)
+
 [H. Park et al.,
 "Mapping inorganic crystal chemical space" _Faraday Discuss._ (2024)](https://pubs.rsc.org/en/content/articlelanding/2024/fd/d4fd00063c)
 

{smact-4.0.0 → smact-4.0.2}/README.md

@@ -1,6 +1,6 @@
 [![DOI](https://joss.theoj.org/papers/10.21105/joss.01361/status.svg)](https://doi.org/10.21105/joss.01361)
-[![DOI](https://zenodo.org/badge/DOI/10.5281/zenodo.5553201.svg)](https://doi.org/10.5281/zenodo.5553201)
-[![Documentation Status](https://readthedocs.org/projects/smact/badge/?version=latest)](http://smact.readthedocs.org/en/latest/?badge=latest)
+[![DOI](https://zenodo.org/badge/DOI/10.5281/zenodo.595853.svg)](https://doi.org/10.5281/zenodo.595853)
+[![Documentation Status](https://readthedocs.org/projects/smact/badge/?version=latest)](https://smact.readthedocs.org/en/latest/?badge=latest)
 [![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://opensource.org/licenses/MIT)
 ![python version](https://img.shields.io/pypi/pyversions/smact)
 [![Ruff](https://img.shields.io/endpoint?url=https://raw.githubusercontent.com/astral-sh/ruff/main/assets/badge/v2.json)](https://github.com/astral-sh/ruff)
@@ -14,7 +14,7 @@
 
 # SMACT
 
-**Semiconducting Materials from Analogy and Chemical Theory** (SMACT) is a collection of rapid screening and informatics tools that uses data about chemical elements.
+**Semiconducting Materials by Analogy and Chemical Theory** (SMACT) is a collection of rapid screening and informatics tools that uses data about chemical elements.
 
 - **Documentation:** <https://smact.readthedocs.io/en/latest/>
 - **Examples:** <https://smact.readthedocs.io/en/latest/examples.html>
@@ -51,6 +51,12 @@ Use cases are available in our [examples](https://smact.readthedocs.io/en/latest
 
 - Compositions can be converted for use in Pymatgen or for representation to machine learning algorithms ([see this example](https://smact.readthedocs.io/en/latest/tutorials/smact_generation_of_solar_oxides.html)) and the related [ElementEmbeddings](https://github.com/WMD-group/ElementEmbeddings) package.
 
+- Charge neutrality screening supports **mixed-valence compounds** via the `mixed_valence=True` flag in `smact_validity`, enabling correct handling of materials like Fe₃O₄ and Mn₃O₄.
+
+- Oxidation state data is sourced from **ICSD 2024**, providing an updated and stricter set of experimentally observed oxidation states per element.
+
+- The [property prediction module](https://smact.readthedocs.io/en/latest/smact.property_prediction.html) enables composition-to-property prediction using pretrained deep learning models, including a ROOST-based band gap predictor trained on the Materials Project database.
+
 - The code also has tools for manipulating common crystal lattice types:
   - Certain structure types can be built using the [builder module](https://smact.readthedocs.io/en/latest/smact.builder.html)
   - Lattice parameters can be estimated using ionic radii of the elements for various common crystal structure types using the [lattice_parameters module](https://smact.readthedocs.io/en/latest/smact.lattice_parameters.html).
@@ -58,6 +64,88 @@ Use cases are available in our [examples](https://smact.readthedocs.io/en/latest
   - The [structure prediction](https://smact.readthedocs.io/en/latest/smact.structure_prediction.html) module can be used to predict the structure of hypothetical compositions using species similarity measures.
   - The [dopant prediction](https://smact.readthedocs.io/en/latest/smact.dopant_prediction.html) module can be used to facilitate high-throughput predictions of p-type and n-type dopants of multicomponent solids.
 
+## Package structure
+
+**Legend:** 🟢 new in v4 — 🟡 improved in v4
+
+```mermaid
+graph TD
+    classDef new fill:#c8e6c9,stroke:#388e3c,color:#000
+    classDef improved fill:#fff9c4,stroke:#f9a825,color:#000
+
+    SMACT(["smact"])
+
+    SMACT --> core["Core modules"]
+    SMACT --> SP["structure_prediction"]
+    SMACT --> DP["dopant_prediction"]
+    SMACT --> PP["🟢 property_prediction"]
+    SMACT --> IO["🟢 io"]
+    SMACT --> UT["utils"]
+
+    core --> init["init.py — Element, Species, neutral_ratios"]
+    core --> dl["data_loader.py — elemental and oxidation state data loading"]
+    core --> sc["🟡 screening.py — compositional screening"]
+    core --> pr["properties.py — band gap, electronegativity, valence electron count"]
+    core --> ox["oxidation_states.py — oxidation state combination likelihood"]
+    core --> mt["metallicity.py — metallic character scoring"]
+    core --> la["lattice.py — Site and Lattice representations"]
+    core --> bld["🟡 builder.py — perovskite and wurtzite structure builders"]
+    core --> lp["🟡 lattice_parameters.py — lattice parameter estimation from ionic radii"]
+    core --> di["distorter.py — inequivalent site enumeration and substitution"]
+
+    sc --> sc1["smact_validity — charge neutrality and Pauling electronegativity test"]
+    sc --> sc2["smact_filter — compositional search space generation"]
+    sc --> sc3["🟢 mixed_valence flag — correct handling of Fe3O4, Mn3O4"]
+    sc --> sc4["🟢 ICSD 2024 oxidation states — stricter, updated elemental data"]
+
+    bld --> bld1["cubic_perovskite — parameterized oxidation state tiling"]
+    bld --> bld2["wurtzite — corrected default cell parameters"]
+
+    lp --> lp1["corrected geometric formulae for all structure types"]
+
+    SP --> spst["structure.py — SmactStructure"]
+    SP --> spdb["database.py — StructureDB SQLite interface"]
+    SP --> spmu["mutation.py — CationMutator from lambda tables"]
+    SP --> sppd["🟡 prediction.py — StructurePredictor"]
+
+    spst --> spst1["from_file, from_mp, from_pymatgen constructors"]
+    sppd --> sppd1["ionic substitution-based crystal structure prediction"]
+    sppd --> sppd2["🟢 updated to mp_api.client.MPRester interface"]
+
+    DP --> doper["🟡 doper.py — Doper"]
+    doper --> doper1["get_dopants — p-type and n-type candidates"]
+    doper --> doper2["to_table — ranked candidates as a formatted table"]
+    doper --> doper3["plot_dopants — periodic table heatmap visualisation"]
+
+    PP --> base["base_predictor.py — BasePropertyPredictor, PredictionResult"]
+    PP --> roost["roost/ — RoostPropertyPredictor"]
+    PP --> conv["convenience.py — predict_band_gap"]
+    PP --> reg["registry.py — model discovery and resolution"]
+
+    roost --> roost1["pretrained ROOST model for band gap prediction"]
+    roost --> roost2["uncertainty estimates alongside predictions"]
+    roost --> roost3["trained on Materials Project 2024 database"]
+    base --> base1["PredictionResult — value, uncertainty, metadata"]
+
+    IO --> ee["🟢 elementembeddings.py — ElementEmbeddings interface"]
+    ee --> ee1["composition_featuriser — composition-level feature vectors"]
+    ee --> ee2["species_featuriser — species-level feature vectors"]
+
+    UT --> comp["composition.py — parse_formula, comp_maker, formula_maker"]
+    UT --> uox["🟢 oxidation.py — ICSD24OxStatesFilter"]
+    UT --> sp2["species.py — parse_spec, unparse_spec"]
+    UT --> cs["crystal_space/"]
+
+    uox --> uox1["consensus and commonality-based filtering of oxidation states"]
+    cs --> cs1["generate_composition_with_smact.py — SMACT-based composition generation"]
+    cs --> cs2["download_compounds_with_mp_api.py — Materials Project bulk download"]
+    cs --> cs3["plot_embedding.py — crystal space visualisation"]
+
+    class PP,IO new
+    class sc,bld,lp,sppd,doper,uox improved
+    class sc3,sc4,sppd2,ee,uox new
+```
+
 ## List of modules
 
 - **smact** library containing:
@@ -75,36 +163,73 @@ Use cases are available in our [examples](https://smact.readthedocs.io/en/latest
     substituting on inequivalent sites of a sub-lattice.
   - **oxidation_states.py**: Used for predicting the likelihood of species coexisting in a compound based on a statistical model.
   - **structure_prediction**: A submodule which contains a collection of tools for facilitating crystal structure predictions via ionic substitutions
-  - **dopant_prediction**: A submodule which contains a collections of tools for predicting dopants.
-  - **utils.py** A collection of utility functions used throughout the codebase.
+  - **dopant_prediction**: A submodule which contains a collection of tools for predicting dopants.
+  - **property_prediction**: A submodule for composition-to-property prediction using pretrained deep learning models (e.g. ROOST band gap predictor).
+  - **utils**: A submodule containing utility functions for composition parsing, species handling, oxidation state filtering, and crystal space generation and download.
 
 ## Requirements
 
 The main language is Python 3 and has been tested using Python 3.11 - 3.13.
-Core dependencies include NumPy, SciPy, pandas, [pymatgen](https://pymatgen.org), [ASE](https://wiki.fysik.dtu.dk/ase), and [spglib](http://atztogo.github.io/spglib). A full list is in [`pyproject.toml`](pyproject.toml).
+Core dependencies include NumPy, SciPy, pandas, [pymatgen](https://pymatgen.org), [ASE](https://wiki.fysik.dtu.dk/ase), and [spglib](https://spglib.readthedocs.io). A full list is in [`pyproject.toml`](pyproject.toml).
 
 ## Installation
 
 The latest stable release can be installed via pip:
 
-    pip install smact
+```bash
+pip install smact
+```
+
+Optional dependencies, needed by most of the examples and tutorials:
+
+```bash
+pip install "smact[optional]"
+```
 
-Optional dependencies (needed for full replication of examples and tutorials):
+The band-gap prediction example additionally needs the pre-trained ROOST models, which
+bring in PyTorch. To run every example and tutorial:
 
-    pip install "smact[optional]"
+```bash
+pip install "smact[optional,property_prediction]"
+```
 
 SMACT is also available via conda-forge:
 
-    conda install -c conda-forge smact
+```bash
+conda install -c conda-forge smact
+```
+
+The optional functionality is available as conda-forge feature packages. Install
+only the feature you need; for example, the pre-trained property-prediction
+models can be installed with:
+
+```bash
+conda install -c conda-forge smact-property-prediction
+```
+
+The other feature packages are `smact-mp`, `smact-crystal-space`,
+`smact-featurisers`, `smact-visualisation`, `smact-ml`, and `smact-optional`.
+`smact-optional` matches the `smact[optional]` pip extra. To install all
+documented examples and tutorials, including property prediction:
+
+```bash
+conda install -c conda-forge smact-optional smact-property-prediction
+```
+
+Conda feature packages use conda-forge builds of compiled dependencies such as
+`pytorch` (the package is named `torch` on PyPI). Select any GPU-specific
+`pytorch` configuration separately for your platform.
 
 ### Developer installation
 
 We use [uv](https://docs.astral.sh/uv/) for dependency management. To set up a development environment:
 
-    git clone https://github.com/wmd-group/smact.git
-    cd smact
-    uv sync --extra optional --extra property_prediction --dev
-    pre-commit install
+```bash
+git clone https://github.com/wmd-group/smact.git
+cd smact
+uv sync --extra optional --extra property_prediction --dev
+pre-commit install
+```
 
 This installs SMACT in editable mode with all optional and development dependencies, and sets up pre-commit hooks. See [CONTRIBUTING.md](CONTRIBUTING.md) for the full workflow.
 
@@ -133,14 +258,21 @@ We use [GitHub Actions](https://github.com/features/actions) for CI. Tests shoul
 
 Run the tests locally:
 
-    make test
+```bash
+make test
+```
 
 Or to run the full CI pipeline (pre-commit hooks and tests):
 
-    make ci-local
+```bash
+make ci-local
+```
 
 ## References
 
+[K. O. Mastej et al.,
+"Chemical filters for ultra-high-throughput materials screening and generation" _arXiv_ (2026)](https://doi.org/10.48550/arXiv.2607.17910)
+
 [H. Park et al.,
 "Mapping inorganic crystal chemical space" _Faraday Discuss._ (2024)](https://pubs.rsc.org/en/content/articlelanding/2024/fd/d4fd00063c)