Single-Cell Sequencing Depth: How Much Is Enough?

Sequencing depth in single-cell RNA-seq (scRNA-seq) is the number of sequencing reads generated per individual cell in an experiment. The question of how much depth is sufficient has no single numeric answer because the optimal depth depends on the biological question, the chosen protocol, the number of cells profiled, and the analytical methods applied downstream. This article provides a calculation framework for estimating required reads per cell based on cell number and biological objective, with practical guidance for researchers planning single-cell experiments.

For researchers planning single-cell experiments, the core decision is a tradeoff between the number of cells sequenced and the depth per cell. Increasing sequencing depth per cell improves gene detection sensitivity up to a saturation point, while increasing cell number improves the resolution of rare cell populations and developmental trajectories. The comparative analysis of six prominent scRNA-seq methods demonstrated that droplet-based methods such as Drop-seq are more cost-efficient for transcriptome quantification of large numbers of cells, while plate-based methods such as Smart-seq2 are more efficient when analyzing fewer cells (Comparative Analysis of Single-Cell RNA Sequencing Methods). This fundamental tradeoff shapes every experimental design decision.

The calculation framework presented here helps researchers estimate sequencing depth requirements by considering the number of cells needed to answer the biological question, the expected gene detection sensitivity of the chosen protocol, and the analytical requirements of downstream applications such as trajectory inference, cluster identification, and variant detection.

Understanding Sequencing Depth in Single-Cell RNA-seq

Sequencing depth in scRNA-seq refers to the number of reads generated per cell after demultiplexing and alignment. This metric differs from bulk RNA-seq depth because each cell represents an independent library with its own capture efficiency, amplification characteristics, and technical noise. The relationship between sequencing depth and data quality is governed by the law of diminishing returns: each additional read per cell increases gene detection until the transcriptome of that cell is saturated, after which additional reads provide minimal new information.

The amount of starting RNA in a single cell is extremely small, and the limited per-cell sequenced reads contribute to technical noise that affects downstream clustering and feature selection (A copula based topology preserving graph convolution network for clustering of single-cell RNA-seq data). This technical noise is a fundamental constraint that cannot be fully eliminated by increasing sequencing depth alone, because the noise originates from the stochastic capture and amplification of individual mRNA molecules.

Single-cell RNA sequencing technologies have enabled comprehensive analysis of immune system heterogeneity by identifying novel distinct immune cell subsets, characterizing stochastic heterogeneity within cell populations, and building developmental trajectories for immune cells (Single-cell RNA sequencing to explore immune cell heterogeneity). Each of these analytical goals has different depth requirements. Identifying major cell types requires less depth than characterizing stochastic heterogeneity within a population, and building developmental trajectories requires sufficient depth to detect the genes that define intermediate states.

The choice of protocol fundamentally determines the depth-sensitivity relationship. Methods that use unique molecular identifiers (UMIs) such as CEL-seq2, Drop-seq, MARS-seq, and SCRB-seq quantify mRNA levels with less amplification noise compared to full-length methods (Comparative Analysis of Single-Cell RNA Sequencing Methods). Full-length methods such as Smart-seq2 and Smart-seq3 detect more genes per cell but require deeper sequencing to achieve comparable quantification accuracy.

Core Principles of Depth Estimation

Gene Detection Saturation

Gene detection saturation is the point at which additional sequencing reads no longer result in the detection of new genes. The saturation curve for a given protocol depends on the library complexity, which is determined by the number of unique mRNA molecules captured and amplified from each cell. Libraries with higher complexity require deeper sequencing to reach saturation, while libraries with lower complexity saturate at lower depths.

Smart-seq3 combines full-length transcriptome coverage with a 5' unique molecular identifier RNA counting strategy that enables in silico reconstruction of thousands of RNA molecules per cell, with 60% of counted and reconstructed molecules directly assigned to allelic origin and 30-50% assigned to specific isoforms (Single-cell RNA counting at allele and isoform resolution using Smart-seq3). This increased sensitivity compared to Smart-seq2 means that Smart-seq3 typically detects thousands more transcripts per cell, which has direct implications for the sequencing depth required to fully utilize the method's capacity.

Cell Number and Rare Population Detection

The number of cells sequenced determines the resolution of rare cell populations. Detecting a rare cell type requires sequencing enough total cells such that the expected number of cells from that population is sufficient for downstream analysis. The mouse organogenesis cell atlas profiled approximately 2 million cells from 61 embryos and identified hundreds of cell types and 56 trajectories, many of which were detected only because of the depth of cellular coverage (The single-cell transcriptional landscape of mammalian organogenesis). This example illustrates that some biological discoveries require massive cell numbers instead of extreme per-cell depth.

For a rare population present at frequency f, the expected number of cells captured is n multiplied by f, where n is the total number of cells sequenced. Detecting a population at 1% frequency with at least 50 cells requires sequencing at least 5,000 cells. Detecting a population at 0.1% frequency with the same cell count requires 50,000 cells. These calculations are independent of sequencing depth per cell, which means that experiments designed to detect rare populations must allocate sequencing resources toward cell number instead of depth.

Biological Question and Analytical Requirements

The biological question determines the minimum depth required for meaningful analysis. Questions that require detecting lowly expressed genes, resolving isoform usage, or identifying somatic variants demand deeper sequencing per cell. Questions that require identifying major cell types or comparing relative proportions across conditions can be answered with shallower depth and larger cell numbers.

The single-cell transcriptional landscape of mammalian organogenesis demonstrated that deep cellular coverage enables the detection of trajectories and cell types that would be missed with smaller datasets (The single-cell transcriptional landscape of mammalian organogenesis). Conversely, studies focused on well-characterized cell types in a tissue may require only enough depth to reliably assign cells to known identities.

At a Glance: Depth Decision Table

The depth ranges in this table are planning estimates based on the comparative performance of different scRNA-seq methods and the analytical requirements of different experimental goals. The comparative analysis of six scRNA-seq methods showed that Smart-seq2 detected the most genes per cell, while UMI-based methods quantified mRNA levels with less amplification noise (Comparative Analysis of Single-Cell RNA Sequencing Methods). These performance differences directly inform the depth requirements for each experimental goal.

Protocol-Specific Depth Requirements

UMI-Based Methods

UMI-based methods including CEL-seq2, Drop-seq, MARS-seq, and SCRB-seq use unique molecular identifiers to correct for amplification bias and enable accurate counting of mRNA molecules. These methods are generally more cost-efficient for transcriptome quantification of large numbers of cells (Comparative Analysis of Single-Cell RNA Sequencing Methods). The UMI counting strategy means that each unique mRNA molecule produces a single count regardless of how many reads are generated from that molecule, which reduces the depth required for accurate quantification.

The practical implication is that UMI-based methods can achieve accurate cell type identification at lower depths than full-length methods. However, UMI-based methods have limited ability to detect isoforms and allelic variants because the reads are typically confined to one end of the transcript.

Full-Length Methods

Full-length methods such as Smart-seq2 and Smart-seq3 provide coverage across the entire transcript, enabling isoform detection, allele-specific expression analysis, and variant calling. Smart-seq3 greatly increased sensitivity compared to Smart-seq2, typically detecting thousands more transcripts per cell (Single-cell RNA counting at allele and isoform resolution using Smart-seq3). The increased sensitivity of Smart-seq3 means that it requires deeper sequencing to fully utilize its capacity for transcript detection.

Long-read single-cell RNA sequencing represents an extension of full-length methods that requires substantially deeper sequencing. A study of ovarian cancer samples increased PacBio sequencing depth to 12,000 reads per cell and captured 152,000 isoforms, of which over 52,000 were novel (Detection of isoforms and genomic alterations by high-throughput full-length single-cell RNA sequencing in ovarian cancer). This depth enabled the detection of cell type-specific isoform usage and gene fusions that were misclassified in matched short-read data.

Spatial Transcriptomic Methods

Spatial transcriptomic methods combine spatial information with single-cell resolution and have different depth considerations than dissociated cell methods. The spatiotemporal transcriptomic atlas of mouse organogenesis used DNA nanoball-patterned arrays and in situ RNA capture to map transcriptional variation during organogenesis with single-cell resolution (Spatiotemporal transcriptomic atlas of mouse organogenesis using DNA nanoball-patterned arrays). These methods face the challenge of balancing resolution, gene capture, and field of view, which affects the effective sequencing depth per spatial location.

Calculation Framework for Estimating Sequencing Depth

Step 1: Define the Biological Question and Required Cell Number

The first step in the calculation framework is to determine the number of cells required to answer the biological question. This determination depends on the expected frequency of the rarest cell population of interest and the minimum number of cells needed from that population for downstream analysis.

For a population present at frequency f, the required total cell number n is calculated as:

n = desired cells from rare population / f

For example, detecting a population at 0.5% frequency with at least 100 cells requires sequencing 20,000 total cells. The mouse organogenesis cell atlas identified hundreds of cell types and 56 trajectories, many detected only because of the depth of cellular coverage from approximately 2 million cells (The single-cell transcriptional landscape of mammalian organogenesis). This example demonstrates that cell number directly determines the resolution of rare populations and trajectories.

Step 2: Determine the Required Depth per Cell

The required depth per cell depends on the analytical goals and the chosen protocol. For cell type identification with UMI-based methods, depths of 10,000-30,000 reads per cell are typically sufficient. For trajectory inference and detection of transiently expressed genes, depths of 30,000-80,000 reads per cell may be required. For isoform and allele resolution with full-length methods, depths of 100,000-500,000 reads per cell are appropriate.

The comparative analysis of scRNA-seq methods showed that power simulations at different sequencing depths revealed different cost-efficiency profiles for different methods (Comparative Analysis of Single-Cell RNA Sequencing Methods). Drop-seq was more cost-efficient for transcriptome quantification of large numbers of cells, while MARS-seq, SCRB-seq, and Smart-seq2 were more efficient when analyzing fewer cells. These findings support the principle that depth per cell should be matched to the number of cells and the biological question.

Step 3: Calculate Total Sequencing Requirement

The total sequencing requirement is calculated as:

Total reads = number of cells x reads per cell

For an experiment with 20,000 cells at 30,000 reads per cell, the total requirement is 600 million reads. For an experiment with 100,000 cells at 20,000 reads per cell, the total requirement is 2 billion reads. These calculations provide the basis for budgeting sequencing resources and selecting the appropriate sequencing platform.

Step 4: Account for Technical Losses

Technical losses during library preparation and sequencing reduce the effective number of reads per cell. These losses include reads that fail quality filtering, reads that align to multiple locations, and reads that are lost during demultiplexing. The ddSeeker tool was developed to perform initial processing and quality metrics of reads generated through Bio-Rad ddSEQ and Illumina experiments, demonstrating that a higher recovery of valid reads is achievable with appropriate processing tools (ddSeeker: a tool for processing Bio-Rad ddSEQ single cell RNA-seq data). Planning should include a buffer of 20-50% additional reads to account for these technical losses.

Step 5: Validate with Pilot Data

A pilot experiment with a small number of cells can provide empirical data on the relationship between sequencing depth and gene detection for the specific protocol and tissue type. The pilot data can be used to generate saturation curves that inform the final depth decision. This validation step is particularly important for non-model organisms or tissues with unusual RNA content, such as the octopus immune cells that required optimized preparation protocols to maintain viability and enable successful library construction (Comprehensive guide for optimizing octopus immune cell preparation to enhance single cell RNA sequencing success).

Practical Implementation Steps

Step 1: Assess Available Resources

Before designing the experiment, assess the available sequencing capacity, budget, and computational resources. The total sequencing requirement calculated from the framework must fit within these constraints. If the calculated requirement exceeds available resources, the experimental design must be adjusted by reducing cell number, reducing depth per cell, or selecting a more cost-efficient protocol.

Step 2: Select the Protocol Based on Biological Question

The protocol selection should be driven by the biological question. For experiments requiring large cell numbers and accurate quantification, UMI-based droplet methods are appropriate. For experiments requiring isoform resolution or allele-specific analysis, full-length methods are necessary. The comparative analysis of six scRNA-seq methods provides a quantitative basis for this choice (Comparative Analysis of Single-Cell RNA Sequencing Methods).

Step 3: Generate Saturation Curves from Pilot Data

A pilot experiment with 500-2,000 cells sequenced at increasing depths can generate saturation curves that show the relationship between reads per cell and genes detected per cell. These curves provide empirical guidance for selecting the depth that balances sensitivity and cost. The saturation point varies by protocol and tissue type, so pilot data from the specific experimental system is valuable.

Step 4: Plan for Downstream Analytical Requirements

Consider the downstream analytical requirements when selecting depth. Trajectory inference requires detection of genes that define intermediate states, which may be expressed at low levels. Variant detection from single-cell transcriptomes requires sufficient depth to detect the variant alleles, and mosaic detection sensitivity is fundamentally constrained by sequencing depth since even the most advanced algorithms cannot identify variants not physically represented in the sequencing library (Strategies for mosaic variant calling in brain disorders). Copy number analysis from single-cell transcriptomes requires sufficient read depth across the genome to estimate copy number profiles at the desired resolution (Delineating copy number and clonal substructure in human tumors from single-cell transcriptomes).

Step 5: Document the Depth Decision

Record the rationale for the selected depth, including the biological question, the expected cell number, the protocol choice, and the pilot data supporting the decision. This documentation supports reproducibility and provides a basis for evaluating whether the depth was sufficient after data analysis.

Records and Measurements

Key Metrics to Track

The following metrics should be recorded for each single-cell sequencing experiment:

Interpreting the Records

The median genes detected per cell and the saturation percentage provide direct evidence about whether the sequencing depth was sufficient. If the saturation percentage is low, additional sequencing may reveal new genes. If the saturation percentage is high, additional sequencing will provide minimal new information and resources would be better allocated to additional cells.

The fraction of reads in cells is a critical quality metric because it indicates the proportion of sequencing reads that contribute to the single-cell data. Low fractions indicate that a substantial portion of the sequencing budget was spent on reads that do not contribute to the analysis, which may indicate problems with library preparation or demultiplexing.

Common Failure Patterns

Insufficient Depth for Rare Cell Detection

A common failure pattern is sequencing too few cells to detect rare populations. The mouse organogenesis cell atlas identified many cell types and trajectories only because of the depth of cellular coverage from approximately 2 million cells (The single-cell transcriptional landscape of mammalian organogenesis). Experiments designed to detect rare populations must allocate sufficient sequencing resources toward cell number.

Excessive Depth Beyond Saturation

Sequencing beyond the saturation point wastes resources that could be allocated to additional cells. The saturation point varies by protocol and tissue type, and pilot data should be used to identify the depth at which additional reads provide minimal new gene detection. The comparative analysis of scRNA-seq methods showed that different methods have different cost-efficiency profiles at different sequencing depths (Comparative Analysis of Single-Cell RNA Sequencing Methods).

Inadequate Depth for Variant Detection

Variant detection from single-cell transcriptomes requires sufficient depth to detect variant alleles. Mosaic detection sensitivity is fundamentally constrained by sequencing depth because variants not physically represented in the sequencing library cannot be identified by any algorithm (Strategies for mosaic variant calling in brain disorders). Experiments designed for variant detection must plan for substantially deeper sequencing than experiments designed only for cell type identification.

Protocol Mismatch with Biological Question

Selecting a protocol that cannot address the biological question is a fundamental design error. UMI-based methods cannot resolve isoforms or alleles, while full-length methods are less cost-efficient for large cell numbers. The comparative analysis of six scRNA-seq methods provides guidance on matching protocols to experimental goals (Comparative Analysis of Single-Cell RNA Sequencing Methods).

Ignoring Technical Noise

Single-cell data is susceptible to technical noise that affects the quality of genes selected for clustering (A copula based topology preserving graph convolution network for clustering of single-cell RNA-seq data). Ignoring this noise can lead to spurious clusters and incorrect biological conclusions. Appropriate normalization and feature selection methods should be applied to account for technical noise.

Limitations and Interpretation Boundaries

Depth Does Not Compensate for Poor Capture

Sequencing depth cannot compensate for poor mRNA capture efficiency. If the library preparation fails to capture the mRNA molecules from a cell, additional sequencing reads will not recover the missing information. The capture efficiency is determined by the protocol and the quality of the input cells.

Gene Detection Is Not Equivalent to Quantification

Detecting a gene with one read is not equivalent to quantifying its expression level accurately. Accurate quantification requires sufficient reads to estimate the number of mRNA molecules, which depends on the UMI counting strategy and the sequencing depth. The comparative analysis of scRNA-seq methods showed that UMI-based methods quantify mRNA levels with less amplification noise (Comparative Analysis of Single-Cell RNA Sequencing Methods).

Depth Requirements Vary by Tissue and Cell Type

Different tissues and cell types have different transcriptome complexities, which affects the depth required for saturation. Cells with high transcriptional diversity require deeper sequencing to achieve the same gene detection sensitivity as cells with lower diversity. Pilot data from the specific tissue type is the most reliable basis for depth decisions.

Computational Methods Cannot Recover Missing Information

No computational method can recover information that was not captured in the sequencing library. The review of mosaic variant calling strategies emphasizes that detection sensitivity is fundamentally constrained by sequencing depth (Strategies for mosaic variant calling in brain disorders). This principle applies broadly to single-cell analysis: the depth decision determines the upper bound of what can be detected.

Quality Controls and Reproducibility

Quality Control Metrics

Quality control for single-cell sequencing experiments should include assessment of cell viability before library preparation, evaluation of library complexity, and monitoring of sequencing quality metrics. The optimized protocol for octopus immune cells demonstrated that maintaining cell viability above 90% for at least 2 hours post-extraction was critical for successful library construction (Comprehensive guide for optimizing octopus immune cell preparation to enhance single cell RNA sequencing success). Similar viability requirements apply to other tissues and cell types.

Reproducibility Assessment

Reproducibility across single-cell RNA-seq protocols for spatial ordering analysis has been examined to understand how protocol choice affects downstream analytical results (Reproducibility across single-cell RNA-seq protocols for spatial ordering analysis). Researchers should assess whether their conclusions are robust to the choice of sequencing depth and protocol by comparing results across conditions.

Data Sharing and Reporting

Reporting the sequencing depth, cell number, and quality metrics in publications supports reproducibility and enables other researchers to evaluate the adequacy of the data. The NIH Genomic Data Sharing Policy provides requirements for data sharing and reporting for NIH-funded research (Genomic Data Sharing Policy). The FAIR Guiding Principles provide a framework for making data findable, accessible, interoperable, and reusable (The FAIR Guiding Principles). The NCBI provides data resources for depositing and accessing genomic data (NCBI Data Resources), and the EMBL-EBI provides training resources for bioinformatics analysis (EMBL-EBI Training).

Professional Escalation Criteria

When to Seek Expert Consultation

Consult a bioinformatics specialist or sequencing core facility when the experimental design involves unusual tissues, non-model organisms, or analytical requirements beyond standard cell type identification. The optimized protocol for octopus immune cells required substantial method development to overcome challenges with cell aggregation, high salinity, and chemical constraints (Comprehensive guide for optimizing octopus immune cell preparation to enhance single cell RNA sequencing success). Similar challenges may arise with other non-model systems.

When to Reconsider the Experimental Design

Reconsider the experimental design when the calculated sequencing requirement exceeds available resources by a substantial margin. Options include reducing the number of cells, selecting a more cost-efficient protocol, or narrowing the biological question. The comparative analysis of scRNA-seq methods provides guidance on cost-efficient protocol selection for different cell numbers (Comparative Analysis of Single-Cell RNA Sequencing Methods).

When to Escalate Data Quality Issues

Escalate data quality issues to the sequencing facility or bioinformatics support when quality metrics fall substantially below expected ranges. Low fractions of reads in cells, low gene detection rates, or poor reproducibility across technical replicates may indicate problems with library preparation, sequencing, or data processing that require expert intervention.

Frequently Asked Questions

What is the difference between sequencing depth and cell number in single-cell RNA-seq?

Sequencing depth refers to the number of reads generated per cell, while cell number refers to the total number of cells profiled in the experiment. These two parameters trade off against each other because the total sequencing budget is finite. Increasing depth per cell improves gene detection sensitivity up to a saturation point, while increasing cell number improves the resolution of rare populations and trajectories. The comparative analysis of scRNA-seq methods showed that different methods have different cost-efficiency profiles depending on whether the experiment prioritizes depth or cell number (Comparative Analysis of Single-Cell RNA Sequencing Methods).

How many reads per cell are needed for basic cell type identification?

For basic cell type identification with UMI-based methods, approximately 10,000-30,000 reads per cell are typically sufficient. This depth enables detection of canonical marker genes that distinguish major cell types. The comparative analysis of scRNA-seq methods showed that UMI-based methods quantify mRNA levels with less amplification noise, which supports accurate cell type identification at moderate depths (Comparative Analysis of Single-Cell RNA Sequencing Methods).

How does the choice of protocol affect sequencing depth requirements?

Protocol choice substantially affects depth requirements. UMI-based methods such as Drop-seq, MARS-seq, and SCRB-seq are more cost-efficient for large numbers of cells, while full-length methods such as Smart-seq2 detect more genes per cell but require deeper sequencing (Comparative Analysis of Single-Cell RNA Sequencing Methods). Smart-seq3 provides increased sensitivity compared to Smart-seq2, detecting thousands more transcripts per cell, which requires deeper sequencing to fully utilize (Single-cell RNA counting at allele and isoform resolution using Smart-seq3).

What depth is needed for trajectory inference in single-cell data?

Trajectory inference requires detection of genes that define intermediate states along developmental or activation trajectories. These genes may be transiently expressed at low levels, requiring deeper sequencing than basic cell type identification. The mouse organogenesis cell atlas identified 56 trajectories, many detected only because of the depth of cellular coverage from approximately 2 million cells (The single-cell transcriptional landscape of mammalian organogenesis). For trajectory inference, depths of 30,000-80,000 reads per cell are typically recommended, with the exact depth depending on the protocol and tissue.

Can increasing sequencing depth compensate for low cell numbers?

Increasing sequencing depth cannot fully compensate for low cell numbers when the goal is to detect rare populations or resolve complex trajectories. The mouse organogenesis cell atlas identified many cell types and trajectories only because of the depth of cellular coverage from approximately 2 million cells (The single-cell transcriptional landscape of mammalian organogenesis). Rare populations present at low frequency require sufficient total cell numbers to capture enough cells from the population of interest, regardless of sequencing depth per cell.

How do I determine the saturation point for my experiment?

The saturation point can be determined empirically from pilot data by sequencing a small number of cells at increasing depths and plotting the relationship between reads per cell and genes detected per cell. The saturation point is the depth at which additional reads provide minimal new gene detection. The comparative analysis of scRNA-seq methods used power simulations at different sequencing depths to evaluate cost-efficiency (Comparative Analysis of Single-Cell RNA Sequencing Methods), and a similar approach can be applied to pilot data from a specific experimental system.

What depth is needed for detecting somatic variants from single-cell transcriptomes?

Detecting somatic variants from single-cell transcriptomes requires substantially deeper sequencing than cell type identification. Mosaic detection sensitivity is fundamentally constrained by sequencing depth because variants not physically represented in the sequencing library cannot be identified by any algorithm (Strategies for mosaic variant calling in brain disorders). For variant detection, depths of 50,000-200,000 reads per cell are typically recommended, with the exact depth depending on the expected variant allele frequency and the analytical methods used.

How should I report sequencing depth in my publication?

Report the median reads per cell, the number of cells passing quality filters, the median genes detected per cell, and the saturation percentage. These metrics enable other researchers to evaluate the adequacy of the sequencing depth and compare results across studies. The NIH Genomic Data Sharing Policy provides requirements for data sharing and reporting for NIH-funded research (Genomic Data Sharing Policy), and the FAIR Guiding Principles provide a framework for making data findable, accessible, interoperable, and reusable (The FAIR Guiding Principles).

Related Bioinformatics Guides

• Single-Cell RNA Sequencing: From Bulk to Resolution
• Master Guide: Single-Cell RNA Sequencing Bioinformatics Workflows
• Single-Cell RNA-Seq Analysis Pipelines for Veterinary Immunology
• Single-cell RNA-seq Trajectory Inference and Cell Lineage Tracing
• Single-Cell RNA-seq Clustering and Cell-Type Annotation Pipelines

References and Further Reading

• EMBL-EBI Training. European Bioinformatics Institute.
• NCBI Data Resources. National Center for Biotechnology Information.
• Genomic Data Sharing Policy. National Institutes of Health.
• The FAIR Guiding Principles. Scientific Data.
• Single-cell RNA sequencing to explore immune cell heterogeneity.. Nature reviews. Immunology, 2018.
• Comparative Analysis of Single-Cell RNA Sequencing Methods.. Molecular cell, 2017.
• Single-cell RNA sequencing of human femoral head in vivo.. Aging, 2021.
• Spatiotemporal transcriptomic atlas of mouse organogenesis using DNA nanoball-patterned arrays.. Cell, 2022.
• The single-cell transcriptional landscape of mammalian organogenesis.. Nature, 2019.
• Single-cell Ribo-seq reveals cell cycle-dependent translational pausing.. Nature, 2021.
• Single-cell RNA counting at allele and isoform resolution using Smart-seq3.. Nature biotechnology, 2020.
• Delineating copy number and clonal substructure in human tumors from single-cell transcriptomes.. Nature biotechnology, 2021.
• Strategies for mosaic variant calling in brain disorders.. 2026.
• PerturbPlan: An analytical framework for designing Perturb-seq experiments. 2026.
• Integrated Analysis of Bulk and Single Cell Transcriptomic Sequencing Reveals Immune Patterns of T Cell Regulation and Prognostic Biomarkers in Glioblastoma. 2026.
• Protocol for identifying cellular reprogramming minimal networks using combinatorial transcription factor screening.. 2026.
• Epcoritamab plus gemcitabine and oxaliplatin in transplant-ineligible relapsed/refractory diffuse large B-cell lymphoma: how should this regimen be positioned in current treatment algorithms?. 2026.
• Construction and validation of a preimplantation kinship identification model using ultra-low-depth whole genome sequencing.. 2026.
• Phylogenetic tree inference from single-cell RNA sequencing data with SCITE-RNA.. 2026.
• Comprehensive guide for optimizing octopus immune cell preparation to enhance single cell RNA sequencing success.. 2026.
• A copula based topology preserving graph convolution network for clustering of single-cell RNA-seq data. bioRxiv, 2021.
• ddSeeker: a tool for processing Bio-Rad ddSEQ single cell RNA-seq data. BMC Genomics, 2018.
• Detection of isoforms and genomic alterations by high-throughput full-length single-cell RNA sequencing in ovarian cancer. bioRxiv, 2023.
• Determining sequencing depth in a single-cell RNA-seq experiment. Nature Communications, 2020.
• Reproducibility across single-cell RNA-seq protocols for spatial ordering analysis. Plos One, 2020.
• An Informative Approach to Single-Cell Sequencing Analysis. Advances in Experimental Medicine and Biology, 2019.

This article is educational and does not replace validated analysis plans, institutional policy, clinical interpretation, or specialist review.