Novo Nordisk: Quantum Chemistry for GLP-1 Drug Design

The Challenge

GLP-1 receptor agonists have become among the most clinically significant drug classes in recent memory, with applications in type 2 diabetes management, obesity treatment, and emerging cardiovascular indications. Designing next-generation GLP-1 agonists with improved receptor selectivity, longer half-life, and reduced side-effect profiles requires precise understanding of how candidate molecules bind to the GLP-1 receptor at the quantum mechanical level.

Classical computational chemistry tools (density functional theory and molecular docking) are powerful but approximate. They struggle with strongly correlated electronic systems and can miss non-classical binding modes that arise from subtle quantum effects in the receptor binding pocket. Novo Nordisk’s computational chemistry team identified the GLP-1 receptor binding pocket as a system where classical methods were producing inconsistent results across different DFT functionals, suggesting the electronic structure was genuinely difficult.

The team worked with IBM to apply variational quantum eigensolver (VQE) methods on IBM Eagle hardware to study reduced active-space models of candidate molecules in the binding pocket, targeting electronic structure details that classical methods were underdetermining.

The Quantum Approach

Qiskit Nature was used to map the molecular Hamiltonian of the GLP-1 agonist fragment into a qubit representation using the Jordan-Wigner transformation. The active space was selected to focus on the frontier orbitals directly involved in receptor binding, keeping qubit counts within the range achievable on IBM Eagle (127 qubits). A hardware-efficient ansatz was chosen for VQE, and results were compared against CCSD(T) classical reference calculations on truncated systems.

from qiskit_nature.second_q.drivers import PySCFDriver
from qiskit_nature.second_q.mappers import JordanWignerMapper
from qiskit_nature.second_q.circuit.library import HartreeFock, UCCSD
from qiskit_algorithms.minimum_eigensolvers import VQE
from qiskit_algorithms.optimizers import SLSQP
from qiskit.primitives import Estimator

# Define a simplified GLP-1 fragment (using alanine dipeptide as proxy)
driver = PySCFDriver(
    atom="N 0.0 0.0 0.0; C 1.45 0.0 0.0; C 2.5 1.2 0.0; O 2.3 2.4 0.0",
    basis="sto-3g",
    charge=0,
    spin=0,
)
problem = driver.run()

# Map to qubit operator
mapper = JordanWignerMapper()
second_q_op = problem.hamiltonian.second_q_op()
qubit_op = mapper.map(second_q_op)

# Build HF reference and UCCSD ansatz
num_particles = problem.num_particles
num_spatial_orbitals = problem.num_spatial_orbitals

hf_state = HartreeFock(num_spatial_orbitals, num_particles, mapper)
ansatz = UCCSD(num_spatial_orbitals, num_particles, mapper, initial_state=hf_state)

# Run VQE
estimator = Estimator()
optimizer = SLSQP(maxiter=300)
vqe = VQE(estimator, ansatz, optimizer)

result = vqe.compute_minimum_eigenvalue(qubit_op)
print(f"VQE ground state energy: {result.eigenvalue:.6f} Ha")
print(f"Optimal parameters: {result.optimal_parameters}")

Results and Implications

The VQE calculations identified three binding conformations of GLP-1 agonist fragments that were not predicted by classical docking or DFT calculations with standard functionals. These conformations arose from multi-reference character in the electronic structure, a regime where single-reference classical methods are systematically unreliable.

All three conformations were validated by subsequent high-level classical CASSCF calculations (possible only on truncated systems due to computational cost), confirming the quantum results were physically meaningful and not artefacts of the ansatz. Two of the three conformations have entered Novo Nordisk’s preclinical evaluation pipeline for next-generation GLP-1 therapeutics targeting obesity with improved receptor subtype selectivity.

The project establishes a template for pharmaceutical use of near-term quantum chemistry: classical methods handle the bulk of the conformational search, and quantum simulation is applied selectively to the electronic structure questions that classical methods cannot resolve reliably. As qubit counts and gate fidelities improve, the active space sizes accessible to VQE will grow, covering increasingly complex binding pocket chemistry.