Quantum-computing-enhanced algorithm unveils potential KRAS inhibitors

Abstract

We introduce a quantum-classical generative model for small-molecule design, specifically targeting KRAS inhibitors for cancer therapy. We apply the method to design, select and synthesize 15 proposed molecules that could notably engage with KRAS for cancer therapy, with two holding promise for future development as inhibitors. This work showcases the potential of quantum computing to generate experimentally validated hits that compare favorably against classical models.

Fig. 1. Schematic representation of the hybrid quantum–classical framework for KRAS ligand development.

a, The initial phase involved assembling a training dataset, starting with 650 experimentally verified KRAS inhibitors sourced from the literature. Using the STONED–SELFIES algorithm, analogs of these inhibitors were generated, expanding the dataset to approximately 850,000 compounds. This was further augmented by adding 250,000 top candidates from a virtual screening of the REAL ligand library against KRAS, creating a total dataset of over 1 million molecules. b, In the generation phase, the dataset was used to train our generative model, consisting of both a classical LSTM network and a QCBM. The LSTM network processed sequential data of chemical structures, while the QCBM, trained on the output from the LSTM, generated complex, high-dimensional probability distributions. This workflow incorporated Chemistry42 as a reward function to encourage the production of structurally diverse and synthesizable molecules. c, Workflow for KRAS inhibitor design, detailing the process from computational compound selection to laboratory synthesis and experimental validation. d, A total of 1 million compounds (classical samples from the LSTM, quantum samples from QCBM on quantum hardware and simulated quantum samples on classical hardware) were evaluated by Chemistry42 to filter out unsuitable candidates and rank the rest by their PLI scores. Finally, 15 promising compounds were selected for synthesis.

Fig. 2. Pharmacological characterization of compounds ISM061-018-2 and ISM061-022 through SPR and cellular activity assays.

a,e, Chemical structures of ISM061-018-2 (a) and ISM061-022 (e). b, SPR sensorgrams illustrating binding effects of ISM061-018-2 on KRAS protein. c,f, MaMTH-DS dose–response curves illustrate the binding kinetics and effects of ISM061-018-2 (c) and ISM061-022 (f) on various KRAS proteins, as well as NRAS, HRAS and artificial bait, with activities measured across concentrations from 4 nM to 30 μM. Data points are presented as averages from n = 4 technical replicates with error bars showing the s.d. Each graph is representative of n = 3 biological replicates, performed under the same conditions. Analyses were carried out using GraphPad Prism. d,g, Results from CellTiter-Glo viability assays measuring the impact of ISM061-018-2 (d) and ISM061-022 (g) on cellular proliferation over a concentration range from 123 nM to 30 μM, demonstrating that they do not display general, nonspecific toxicity. Data points are presented as averages from n = 3 technical replicates with error bars showing the s.d. Each graph is representative of n = 3 biological replicates performed under the same conditions.

Extended Data Fig. 1. Comparative Benchmarking of Quantum and Classical Ligand Design Methods.

(A, B) Comparative analysis of our hybrid approaches with varied priors. The performance of the Quantum Circuit Born Machine (QCBM) was assessed using both a quantum simulator (Sim) and a hardware backend (HW), and contrasted with a Multi-bases QCBM (MQCBM) operating solely on a quantum simulator (Sim), as well as an LSTM model devoid of quantum priors (representing a fully classical architecture). We calculated the number of generated molecules that met a series of synthesizability and stability criteria as stipulated by the Tartarus benchmarking platform (referred to as Local Filters) and by Chemistry42 (referred to as Chemistry42 Filters). To generate Figure A, we repeated the experiments five times, sampling n=1,000 compounds in each repetition and applying the filter. The reported values represent the mean for each data point, with error bars indicating the standard deviation across the repetitions (Mean ± Std.). more detailed is reported in Supplementary Information Table S3.2. (C) Success rate of generating molecules that meet Tartarus’s filter criteria as a function of the number of qubits used in modeling priors for the QCBM. We Comparative Benchmarking of Quantum and Classical Ligand Design Methods. (A, B) Comparative analysis of our hybrid approaches with varied priors. The performance of the Quantum Circuit Born Machine (QCBM) was assessed using both a quantum simulator (Sim) and a hardware backend (HW), and contrasted with a Multi-bases QCBM (MQCBM) operating solely on a quantum simulator (Sim), as well as an LSTM model devoid of quantum priors (representing a fully classical architecture). We calculated the number of generated molecules that met a series of synthesizability and stability criteria as stipulated by the Tartarus benchmarking platform (referred to as Local Filters) and by Chemistry42 (referred to as Chemistry42 Filters). To generate Figure A, we repeated the experiments five times, sampling n=1,000 compounds in each repetition and applying the filter. The reported values represent the mean for each data point, with error bars indicating the standard deviation across the repetitions (Mean ± Std.). more detailed is reported in Supplementary Information Table S3.2. (C) Success rate of generating molecules that meet Tartarus’s filter criteria as a function of the number of qubits used in modeling priors for the QCBM. We repeated the experiments five times, sampling n=5,000 compounds in each repetition and applying the filter. The reported values represent the mean for each data point, with error bars indicating the standard deviation across the repetitions.

Extended Data Fig. 2. Protein-detected NMR experiments.

Protein-detected NMR experiments elucidate the binding mode of Compound 18-2 to KRAS-G12D. (A) Significant chemical shift perturbations (CSPs) and intensity changes due to chemical exchange are observed upon binding to the compound. Residues lining the Switch-II pocket that show significant CSPs are highlighted in the inset, including Gly10 (from the P-loop), Phe78 and Gly77 (from the α-2 helix adjacent to the Switch-II region), Ser89, and Asp92 (from the β-4 sheet). (B) The CSP values are mapped onto the G12D structure and displayed in a color gradient. Residues that could not be assigned are shown in grey. The majority of significant CSPs are observed near the Switch-II pocket; however, CSPs are also noted in the α-1 helix, Switch-I, and α-4 regions, possibly indicating conformational changes upon binding to the compound. (C) Zoomed-in region of the protein, highlighting residues from the Switch-II pocket that exhibit CSPs.

Extended Data Fig. 3. Quantum-Enhanced Generative Model for Drug Discovery Applications.

(B) Hybrid model combining a Quantum Circuit Born Machine (QCBM) with Long Short-Term Memory (LSTM). This model iteratively trains using prior samples from quantum hardware. (A) Integration method of prior samples into the LSTM architecture. Molecular information (in SELFIES encoding) and quantum data are merged by addition or concatenation. The resultant samples, X’(t), are then input to the LSTM cell. (C) Quantum prior component described as a QCBM, generating samples from quantum hardware each training epoch and trains with a reward value, P (x)=Softmax(R(x)), calculated using Chemistry42 or a local filter.

Extended Data Fig. 4. Pharmacophore Model Depiction.

Pharmacophore Model Depiction for KRAS Inhibitor TH-Z816 Based on the Co-crystallized Ligand Structure Analyzed with Chemistry42 (PDB: 7EW9). This figure illustrates the key pharmacophoric features identified from the ligand structure of TH-Z816 bound to KRAS G12D. A blue sphere represents a critical ring system that supports structural integrity, a green sphere highlights a hydrophobic moiety essential for binding affinity, and a cyan sphere indicates a hydrogen bond donor that contributes to interaction specificity with the KRAS protein. The protein structure is shown on the right, while the pharmacophore interactions within the KRAS Switch-II binding pocket are detailed on the left.

Extended Data Fig. 5. Quantum Circuit Born Machine (QCBM) Model.

Schematic representation of the Quantum Circuit Born Machine (QCBM) implemented in our numerical experiments, illustrating a variational quantum circuit with a configuration of three layers and four qubits. In practice, our numerical experiments utilized a system with 16 qubits. The depicted quantum gates, including parameterized rotations (Rx, Rz) and entangling CNOT gates, are orchestrated to evolve the initial state $∣0⟩$ into a complex quantum state $∣\mathrm{\Psi }\left(0\right)⟩$. The outcome is measured, and the resulting data are used by the classical optimizer to iteratively refine the parameters θ, thus leading the circuit towards an optimal solution for ligand generation.