CARE: Catalysis Automated Reaction Evaluator

CARE (Catalytic Automated Reaction Evaluator) is a framework for the automated generation and manipulation of chemical reaction networks (CRNs) in heterogeneous catalysis. CARE is powered by ML-based energy evaluators (GAME-Net-UQ, FairChem, MACE, UPET, Orb, SevenNet) and includes multiscale kinetic functionalities enabling the quantification of catalytic activity for reactions containing thousands of elementary steps.

🪛 Installation

To install CARE with a specific Machine Learning Interatomic Potential (MLIP) evaluator, specify it as an extra dependency.

pip install "care-crn[mlip]"

Replace [mlip] with your desired evaluator. Supported models include: fairchemv1 | fairchemv2 | mace | upet | orb | sevenn | gamenetuq

[!WARNING] Environment Isolation Because each ML model depends on highly specific backend versions (PyTorch, e3nn, ASE, etc.), you will likely need to create one distinct virtual environment for each ML evaluator you intend to use to avoid dependency conflicts.

[!NOTE] Automatic Julia Backend Setup CARE relies on a Julia backend for high-performance ODE integration during microkinetic simulations. No manual installation of Julia is required.

The first time you execute a simulation, juliapkg will automatically download a private, compatible version of Julia (if not present) and install the necessary dependencies defined in src/care/juliapkg.json into an isolated environment. This initial setup will take a few extra minutes to compile.

Developer Installation

Required disk space: ~6.5 GB (Python environment), ~4.3 GB (Julia+dependencies)

1. Clone the repo:

   git clone git@github.com:LopezGroup-ICIQ/care.git
   cd care
   

2. Create environment:

   conda create -n care_env python==3.12
   conda activate care_env
   

3. Install care-crn in editable mode:

   python3 -m pip install -e .[gamenetuq,mace,etc.]
   

💥 Usage

Network Generation

You can construct a reaction network blueprint using one of three primary methods:

1. Reactants and Products: Define the start and end points using SMILES strings.
2. Cutoffs: Define the maximum number of Carbon (ncc) and Oxygen (noc) atoms allowed.
3. Chemical Space: Provide a specific list of target SMILES strings to explore (e.g., a specific decomposition network).

[!TIP] Supported Chemical Space CARE currently supports reaction networks with species containing CHONS + Halogens.

from care import ReactionNetwork

# Method 1: From explicit reactants and products (e.g., CO2 hydrogenation to Methanol)
crn = ReactionNetwork.from_species(reactants=["O=C=O", "[H][H]"], products=["CO", "O"])

# Method 2: From elemental network cutoffs (e.g., max 2 Carbons, 1 Oxygen)
crn = ReactionNetwork.from_cutoffs(ncc=2, noc=1)

# Method 3: From a predefined chemical space (e.g., Ethanol decomposition)
crn = ReactionNetwork.from_chemical_space(cs=["CCO"])

# Prints a quick summary (number of species, reactions, etc.)
print(crn) 

# Returns a pandas DataFrame of the generated reactions
df_reactions = crn.get_reaction_table() 

# Renders a visual graph of the reaction network
crn.plot() 

Energy Evaluation

The range of catalyst materials on which CRNs can be evaluated depends on the training domain of the employed ML model.

[!NOTE] Available Evaluators A complete list of available ML evaluators (and their specific capabilities) can be found in the Evaluators README.

from care import Surface 
from care.evaluators import MACEevaluator

# 1. Define the catalyst surface (e.g., Pt(110) from the Materials Project)
surface = Surface.from_mp("mp-2", mp_api_key="your_key", hkl="110", xy_repeat=2)
crn.add_catalyst(surface)

# 2. Initialize the ML evaluator (using GPU acceleration if available)
ml_evaluator = MACEevaluator(device="cuda", num_configs=3, max_steps=50, fmax=0.05)

# 3. Relax all adsorbed intermediates
for intermediate in crn.intermediates.values():
    ml_evaluator(intermediate)

# 4. Perform Nudged Elastic Band (NEB) for transition state searches
for reaction in crn.reactions:
    ml_evaluator(reaction, num_images=3)

# View the updated table containing the newly computed thermodynamics and barriers
crn.get_reaction_table()

Microkinetic Run

Once the CRN is energetically evaluated, microkinetic simulations enable you to quantify the performance of the catalyst. CARE automatically calculates apparent kinetics ($E_{app}$, $n_{app}$) and performs robust sensitivity analyses, including the Degree of Rate Control ($\chi_{RC}$) and Degree of Selectivity Control ($\chi_{SC}$).

from care.reactors import DifferentialPFR

# Initialize the reactor model
reactor = DifferentialPFR(crn)

# Define operating conditions
reactor.T = 473  # Temperature in K
reactor.P = 1e6  # Pressure in Pa
y0 = {"CO2": 0.2, "H2": 0.4, "Ar": 0.4}  # Inlet gas molar fractions
cov0 = {"H": 0.9}                        # Initial surface coverages

# Execute the simulation (automatically delegates to the Julia backend)
mkm = reactor.run(iv=y0, cov0=cov0, eapp=True, napp=True, drc=True)

# View results directly in the terminal
print(mkm.get_performance_summary())

# Export comprehensive results to Excel
mkm.export("mkm_report.xlsx")

Run all together

You can run the entire pipeline (blueprint generation ➡ energy evaluation ➡ kinetic simulation) running the care_run script:

care_run -h  # documentation
care_run -i input.toml -o output_name

This will generate a output_name folder with the generated reaction network and additional results from the kinetic simulation. Examples of input .toml files can be found here.

📖 Tutorials

We currently provide two tutorials, available in the notebooks directory:

• CARE tutorial (run it on Google Colab by clicking the badge on top of the README)
  
• Adsorbate placement

✒️ License

The code is released under the MIT license.

📜 Reference

If you use CARE in your research, please cite the following paper and consider starring the repository.

Morandi, S., Loveday, O., Renningholtz, T. et al. An end-to-end framework for reactivity in heterogeneous catalysis. Nat. Chem. Eng. (2026). https://doi.org/10.1038/s44286-026-00361-8

@article{CARE,
  title = {An End-to-End Framework for Reactivity in Heterogeneous Catalysis},
  author = {Morandi, Santiago and Loveday, Oliver and Renningholtz, Tim and {Pablo-Garc{\'i}a}, Sergio and {Vargas-Hern{\'a}ndez}, Rodrigo A. and Seemakurthi, Ranga Rohit and Sanz Berman, Pol and {Garc{\'i}a-Muelas}, Rodrigo and {Aspuru-Guzik}, Al{\'a}n and L{\'o}pez, N{\'u}ria},
  year = {2026},
  journal = {Nature Chemical Engineering},
  volume = {3},
  number = {3},
  pages = {169--180},
  doi = {10.1038/s44286-026-00361-8},
}