Quantum Generative Modeling of Single-Cell transcriptomes: Capturing Gene-Gene and Cell-Cell Interactions

TL;DR

qSimCells introduces a quantum kernel-based generator that uses entanglement in a parameterized quantum circuit to jointly model intra-cellular gene regulation and inter-cellular ligand–receptor communication, producing realistic, heterogeneous scRNA-seq data. By sampling a final entangled state and applying a Negative Binomial augmentation, it creates ground-truth datasets with non-classical dependencies that conventional correlation-based methods struggle to recover, enabling robust benchmarking of inference algorithms. The study demonstrates that programmed CX-based causal paths yield distinct expression patterns and cell-type separations, while classical GRN inference misinterprets these as spurious correlations; CellChat shows relative increases in true LR signals under inter-state interaction, validating the mechanistic links. Overall, qSimCells argues for quantum-native generative modeling in single-cell biology to capture nonlinear, directional dependencies beyond the reach of classical methods, providing a new benchmark and a path toward quantum-aware inference techniques.

Abstract

Single-cell RNA sequencing (scRNA-seq) data simulation is limited by classical methods that rely on linear correlations, failing to capture the intrinsic, nonlinear dependencies. No existing simulator jointly models gene-gene and cell-cell interactions. We introduce qSimCells, a novel quantum computing-based simulator that employs entanglement to model intra- and inter-cellular interactions, generating realistic single-cell transcriptomes with cellular heterogeneity. The core innovation is a quantum kernel that uses a parameterized quantum circuit with CNOT gates to encode complex, nonlinear gene regulatory network (GRN) as well as cell-cell communication topologies with explicit causal directionality. The resulting synthetic data exhibits non-classical dependencies: standard correlation-based analyses (Pearson and Spearman) fail to recover the programmed causal pathways and instead report spurious associations driven by high baseline gene-expression probabilities. Furthermore, applying cell-cell communication detection to the simulated data validates the true mechanistic links, revealing a robust, up to 75-fold relative increase in inferred communication probability only when quantum entanglement is active. These results demonstrate that the quantum kernel is essential for producing high-fidelity ground-truth datasets and highlight the need for advanced inference techniques to capture the complex, non-classical dependencies inherent in gene regulation.

Figures (4)

• Figure 1: Quantum-simulated single-cell data framework. Our framework utilizes a quantum kernel, divided into three key sections, to generate realistic single-cell data. First, independent feature (gene) activation is simulated for two distinct cell types (CT1 and CT2). Second, a gene regu- latory network (GRN) is established using controlled-NOT (CX) gates between source and target qubits, modeling gene-gene interactions. Third, inter-cellular (inter-state) communication channels are introduced and enhanced by CX gates, representing interactions between cell types. The simulation generates sampled binary histograms for CT1 and CT2, encoding complex inter- and intra-cellular communication within the feature (qubit) states. These histograms are then converted into a binary matrix, where each entry represents the simultaneous activation of features (genes) within individual simulated cells. The final step involves transforming this binary matrix into gene expression profile using a negative binomial augmentation, thus creating a more biologically realistic synthetic single-cell data.
• Figure 2: Quantum circuit sampling. A) shows inter-state interactions modeled by $L_1 = \{ (3, 5), (5, 7), (7, 0) \}$ entanglement interactions on the quantum circuit, and the corresponding measurements per $\ket{\psi_0}$ state. B) shows non-communicating intra-state modeled by $L_2 = \{ (2, 1) \}$ entanglement on the quantum circuit, and the corresponding measurements per $\ket{\psi_0}$ state.
• Figure 3: UMAP visualization of simulated scRNA-seq data. The UMAP plot displays the synthetic single-cell data generated under two different entanglement scenarios (Case 1: Inter-state cascade; Case 2: Non-interacting control), showing the resulting lineage separation between CT1 and CT2 populations.
• Figure 4: Gene regulatory networks through classical prediction. A) shows GRN inference through Pearson and Spearman correlation computed adjacency matrix above 0.5 threshold for inter-state interaction simulation. B) shows GRN inference through Pearson and Spearman correlation computed adjacency matrix above 0.5 threshold for non-inter-state simulation.