TL;DR
This paper introduces miRTalk, a novel computational framework for inferring cell-cell communication mediated by extracellular vesicle-derived microRNAs (EV-miRNAs) using single-cell RNA sequencing (scRNA-seq) data. miRTalk addresses a critical gap in understanding how these EV-miRNAs influence cellular interactions, offering a robust and accurate method to uncover these complex biological processes.
What Problem Does This Paper Address?
Cell-cell communication (CCC) is fundamental to multicellular life, and extracellular vesicles (EVs) carrying microRNAs (miRNAs) have emerged as a key mechanism for this communication. These EV-miRNAs can travel between cells and inhibit gene expression, thereby altering cellular behavior. While single-cell RNA sequencing (scRNA-seq) has revolutionized our ability to study cellular heterogeneity, inferring EV-miRNA-mediated CCC from this data remains a significant computational challenge due to the lack of specialized tools. This paper aims to fill that gap by providing a dedicated framework, miRTalk, to address this specific type of intercellular communication.
Methods
The core of miRTalk is a probabilistic model designed to infer EV-miRNA-mediated CCC. The model operates on the principle that CCC occurs when a miRNA is highly expressed in sender cells and its target genes are lowly expressed in receiver cells. To support this, the authors built miRTalkDB, a comprehensive database of experimentally verified EV-derived miRNA-target associations for humans, mice, and rats. This database was curated by integrating information from existing repositories like ExoCarta, Vesiclepedia, and EVmiRNA, along with manual literature mining.
The algorithm involves:
- Filtering widely expressed miRNAs and highly variable target genes in sender and receiver cells.
- Pairwise sampling to identify sender-receiver pairs exhibiting elevated miRNA expression in senders and downregulated target gene expression in receivers.
- A probabilistic model to calculate the probability of miRNA-target interactions, considering both miRNA expression in sender cells and target gene expression in receiver cells.
- Aggregation of miRNA activity to account for multiple miRNA genes encoding the same mature miRNA.
- Integration of miRNA-target interaction scores and probabilities to derive a final score for EV-miRNA-mediated CCC.
The tool also includes visualization capabilities using chord, circle, bubble, and heatmap plots to illustrate CCC networks from multiple perspectives.
Datasets
The study utilized several scRNA-seq datasets for training and evaluation:
- Simulated scRNA-seq data: Generated using the ESCO R package to create benchmark datasets with controlled parameters (gene number, cell number, miRNA/target coverage, sender/receiver ratio) for robust performance evaluation.
- Real-world datasets:
- Human cholangiocarcinoma (GEO: GSE138709)
- Human bladder cancer, glioblastoma multiforme, and ovarian cancer (TISCH2 database: GSE130001, GSE84465, GSE165897)
- Murine kidney fibrosis (Zenodo: https://zenodo.org/record/4059315)
- Normal and fatty liver transplantation (GSA: CRA004061)
- Human glioblastoma multiforme (TCGA and CPTAC)
The paper also references existing ligand-receptor (LR)-based CCC inference methods (CellPhoneDB, CellChat, NICHES, ICELLNET) for comparative benchmarking.
Key Insights and Ideas
- EV-miRNAs are a significant, yet understudied, mode of CCC: The paper highlights that while EV-miRNAs are known to mediate CCC, computational tools for their inference from scRNA-seq data were lacking. miRTalk fills this crucial gap.
- The importance of curated databases: The development of miRTalkDB, a comprehensive and experimentally verified database of EV-miRNA-target associations, is central to the method's accuracy and novelty.
- miRTalk's probabilistic approach captures inhibitory effects: Unlike some LR-based methods that focus on positive correlations, miRTalk's model specifically accounts for the inhibitory nature of miRNA-target interactions.
- Disease-specific CCC mechanisms are complex and can be uncovered: The application of miRTalk to disease datasets revealed intricate EV-miRNA-mediated CCC networks that were previously unseen, offering insights into disease pathogenesis.
Key Results
- High accuracy and robustness: miRTalk demonstrated remarkable accuracy on both simulated and real-world datasets, achieving high AUROC and AUPRC scores (e.g., 0.93 and 0.98 on simulated data).
- Performance comparable to LR-based methods: When adapted to LR inference, miRTalk showed comparable performance to established methods.
- Uncovering disease-specific communication: The tool successfully identified EV-miRNA-mediated CCC in various disease contexts, including bladder cancer, cholangiocarcinoma, ovarian cancer, glioblastoma, kidney fibrosis, and liver transplantation.
- Identification of key miRNAs in disease: The study highlighted specific miRNAs (e.g., miR-24-3p, miR-142-3p, miR-125b-5p) and their roles in disease progression and cellular interactions. For instance, miR-125b-5p from malignant GBM cells was shown to interact with PABPC1 in recipient cells, influencing metabolic processes and survival.
Limitations and Future Work
- miRNA coverage in scRNA-seq data: The paper acknowledges that the coverage of miRNAs in current scRNA-seq datasets can be sparse. However, it notes that ongoing advancements in sequencing technologies (e.g., snRandom-seq) and computational frameworks (e.g., PPMS) are improving miRNA detection, which will further enhance miRTalk's utility.
- Database expansion: While miRTalkDB is comprehensive, future work could involve expanding it to include more species and a wider range of EV-derived miRNA-target interactions.
- Integration with other CCC modalities: The authors suggest that miRTalk could be integrated with methods inferring other forms of CCC (e.g., ligand-receptor interactions, metabolic signaling) for a more holistic understanding of intercellular communication.
- Further validation in diverse biological contexts: While validated on several disease datasets, applying miRTalk to a broader range of physiological and pathological conditions would further solidify its applicability.
Summary updated