Direct Air Capture¶

Predict CO₂ adsorption isotherms for metal-organic frameworks (MOFs) using ALIGNN trained on the hypothetical MOF (hMOF) dataset. Outputs CO₂ uptake at 5 pressure points (0.01, 0.05, 0.1, 0.5, 2.5 bar) from a single POSCAR input. The model (hmof_co2_absp_alignn) is auto-downloaded from figshare on first use.

Open App

Overview¶

The DAC app uses a specialized ALIGNN model with 5 output features, each corresponding to CO₂ adsorption (mol/kg) at a different pressure point. The model was trained on the hMOF dataset of ~130K hypothetical metal-organic frameworks with GCMC-computed CO₂ isotherms. Input any MOF crystal structure as POSCAR and get the predicted isotherm in seconds.

Data Source

ALIGNN model hmof_co2_absp_alignn — auto-downloaded from figshare via alignn.pretrained.get_figshare_model(). Trained on the hMOF dataset (hypothetical MOFs with GCMC CO₂ isotherms).

Endpoints¶

`POST /dac/predict` — Predict CO₂ isotherm from POSCAR¶

curl -X POST "https://atomgpt.org/dac/predict" \
  -H "Authorization: Bearer sk-XYZ" \
  -H "Content-Type: application/json" \
  -H "accept: application/json" \
  -d '{
    "poscar": "ZIF-8\n1.0\n-8.496 8.496 8.496\n8.496 -8.496 8.496\n8.496 8.496 -8.496\nZn N H C\n6 24 96 48\ndirect\n0.75 0.25 0.5\n0.25 0.75 0.5\n0.5 0.75 0.25\n0.5 0.25 0.75\n0.25 0.5 0.75\n0.75 0.5 0.25\n0.651 0.093 0.378\n0.714 0.273 0.622\n0.286 0.907 0.559\n0.349 0.727 0.441\n0.378 0.651 0.093\n0.622 0.714 0.273\n0.559 0.286 0.907\n0.441 0.349 0.727\n0.093 0.378 0.651\n0.273 0.622 0.714\n0.907 0.559 0.286\n0.727 0.441 0.349\n0.093 0.651 0.378\n0.273 0.714 0.622\n0.907 0.286 0.559\n0.727 0.349 0.441\n0.651 0.378 0.093\n0.714 0.622 0.273\n0.286 0.559 0.907\n0.349 0.441 0.727\n0.378 0.093 0.651\n0.622 0.273 0.714\n0.559 0.907 0.286\n0.441 0.727 0.349"
  }'

Field	Type	Default	Description
`poscar`	string	required	VASP POSCAR of MOF structure (max 50KB)

Response:

Field	Description
`pressures`	`[0.01, 0.05, 0.1, 0.5, 2.5]` (bar)
`adsorption`	Predicted CO₂ uptake at each pressure (mol/kg)
`pressure_unit`	`"bar"`
`adsorption_unit`	`"mol/kg"`
`formula`	Reduced chemical formula
`num_atoms`	Number of atoms
`spacegroup`	Space group

Python Examples¶

Predict isothermPlot isothermScreen multiple MOFs

import requests

ZIF8 = open("ZIF-8.vasp").read()

response = requests.post(
    "https://atomgpt.org/dac/predict",
    headers={
        "Authorization": "Bearer sk-XYZ",
        "accept": "application/json",
        "Content-Type": "application/json",
    },
    json={"poscar": ZIF8},
)
data = response.json()
if data["success"]:
    print(f"{data['formula']} ({data['num_atoms']} atoms)")
    for p, a in zip(data["pressures"], data["adsorption"]):
        print(f"  {p:>5.2f} bar → {a:.5f} mol/kg")

import requests
import matplotlib.pyplot as plt

response = requests.post(
    "https://atomgpt.org/dac/predict",
    headers={
        "Authorization": "Bearer sk-XYZ",
        "accept": "application/json",
        "Content-Type": "application/json",
    },
    json={"poscar": open("MOF.vasp").read()},
)
data = response.json()

plt.figure(figsize=(7, 5))
plt.semilogx(data["pressures"], data["adsorption"], "o-", color="#22c55e", lw=2, ms=8)
plt.xlabel("Pressure (bar)")
plt.ylabel("CO₂ Adsorption (mol/kg)")
plt.title(f"{data['formula']} — CO₂ Isotherm (ALIGNN)")
plt.grid(alpha=0.3)
plt.tight_layout()
plt.savefig("co2_isotherm.png")

import requests
import glob

results = []
for f in glob.glob("mofs/*.vasp"):
    response = requests.post(
        "https://atomgpt.org/dac/predict",
        headers={
            "Authorization": "Bearer sk-XYZ",
            "accept": "application/json",
            "Content-Type": "application/json",
        },
        json={"poscar": open(f).read()},
    )
    data = response.json()
    if data["success"]:
        # Adsorption at 0.5 bar (index 3)
        results.append((data["formula"], data["adsorption"][3]))

results.sort(key=lambda x: x[1], reverse=True)
print("Top MOFs by CO₂ uptake at 0.5 bar:")
for formula, ads in results[:10]:
    print(f"  {formula}: {ads:.4f} mol/kg")

AGAPI Agent¶

from agapi.agents import AGAPIAgent
import os

agent = AGAPIAgent(api_key=os.environ.get("AGAPI_KEY"))

# Predict isotherm
response = agent.query_sync("Predict the CO2 adsorption isotherm for ZIF-8")
print(response)

References¶

Kamal Choudhary, Taner Yildirim, Daniel W Siderius, A Gilad Kusne, Austin McDannald, Diana L Ortiz-Montalvo, Comp. Mat. Sci. 210, 111388 (2022) DOI
K. Choudhary, Comp. Mat. Sci. 259, 114063 (2025) DOI
K. Choudhary, npj Comp. Mat. 7, 185 (2021) — ALIGNN DOI
atomgptlab/alignn