A perspective on quantum computing for analyzing cell-cell communication networks

Abstract

Cell-cell communication enables multicellular life by facilitating interactions between cells, coordinate activities, and performing biological functions. Disentangling these intricate interactions for understanding disease and tissue function is important to reveal the biological underpinnings of complex biological processes such as tumor progression. Recent advances in single-cell RNA sequencing (scRNA-seq) have allowed the development of several computational tools to infer cell-cell interactions (CCI) through integrating gene expression data along with ligand-receptor interactions (LRI). CCI can be predicted using gene expression matrix filtered by ligands and receptors. Different approaches such as co-expression analysis, ligand-receptor pairings, pathway analysis are employed to create networks of CCI, which are then analyzed and visualized to explore communication patterns across the cellular landscape. These methods are often validated using protein-protein interactions and spatial mapping information. Single cell trajectory and kinetic analysis have been the standard practice when looking at the relationship between different single cell sub-populations and often used to identify cell-cell pairs with varying communicative potential between samples. One of the many avenues of the study of cell-state transitions is to infer key drivers of drug resistance in gene regulatory networks as well as possible drug tolerance. This might lead to a more informed understanding of trajectories and heterogeneity of cancer stem cells for biomarker discovery. Furthermore, studying therapeutic resistance and disease progression is significant to counter transformations to high-grade malignancy. Indeed, diseases such as Richter syndrome arising from chronic lymphocytic leukemia is a classic example of such transformation and although there has been prior work in understanding key genetic drivers of transformations to high-grade malignancy, one can use quantum computing applications in network science to study more nuanced outcomes and improve cell-free DNA analyses for prognostic and diagnostic tasks in cancer research. There exist several unsupervised methods which detect changes in gene expression or in graph embeddings to construct single cell trajectories (e.g. SEURAT MONOCLE SCANPY, Celltalker, CellPhoneDB, etc.). In general, all the mentioned methods enable the display of dynamics and characteristics of transition between different cell states using LRI. Other approaches for CCI identification use network-based methods such as NicheNet, CCCExplorer, SpaOTsc, etc. They either use statistical and optimization strategies to infer CCC or measure predictive power of the LRI network using downstream pathway target. CellChat, a popular tool in inferring CCI, uses unsupervised clustering, forms a CCC network and uses a random walk-based network propagation technique to validate the interactions. Thus, classical approaches are often rendered as clustering cells into subgroups and order them along a continuous path that indicate the evolution of the process of interest. However, this offers its own challenge for analyzing subpopulation confirmation and its continuous transition along trajectories is not a trivial task due to computational cost, cell-to-cell variation and heterogeneity as well as the amount of available data. The current global approach to measuring distances between gene distributions or cell type pairs may not accurately reflect local interactions within tissue structures, leading to spurious CCI information. Several other challenges exist in processing scRNA-seq data for CCI such as quantifying uncertainty of measurements and analysis, validating and benchmarking these tools, integration of single-cell data across samples and experiments, among others. The complex CCC networks and the challenges involved in processing them offer an intriguing prospect for quantum computing (QC). Recent advances in QC, such as quantum machine learning (QML) methods have shown to be promising in various fields of interests including network analysis as well as cell-centric therapeutic development. Quantum algorithms make use of a radically different paradigm to learn meaningful CCI from dense networks of LRI and other modalities of data involving spatial and temporal information. Here, we propose two such approaches in processing scRNA-seq data to form networks of LRI and employ quantum equivalents of random walk and graph neural networks (GNN) to infer and analyze CCIs. By mapping high-dimensional spatial transcriptomics data to a lower-dimensional quantum state space, quantum Born machine (QBM) may help in simplifying the representation of tissue structures, which could facilitate more effective distance measurements. Similar to classical machine learning techniques like Principal Component Analysis (PCA), QBM can reduce the dimensionality of the data while preserving essential information. Identification of cell groups with similar gene expression patterns involves applying a graph-based clustering method to a combined dataset of tumor and normal cells. Utilizing the principles of quantum mechanics such as superposition and interference, quantum walks can explore complex cellular networks efficiently and potentially reveal intricate patterns of gene expression that classical methods might miss. However, given the current stage of quantum computing, one can use a hybrid approach that exploits quantum computing for initial data exploration and classical methods for detailed analysis and validation. The use of hybrid quantum approach can then help shed some lights to complex CCC in therapeutic resistance and disease progression.