Chapter Four · failure evidence
What Single-Cell Sequencing got wrong, from 65 dissertations
The compiled records demonstrate technical, statistical, and experimental hurdles encountered when applying single-cell sequencing methodologies across diverse biological tissues. Researchers frequently experienced difficulties including transcript dropout, dissociation artifacts, pseudoreplication in differential expression, and batch integration errors. These records come from PhD theses at 15 institutions, 2021 to 2026. Each links to its thesis. They were extracted by language models reading the full text, so treat each as a lead to read, not a verdict.
Shallow sequencing depth and transcript dropout obscure molecular targets
Low per-cell read coverage and allelic dropout led to missed transcripts and false positive somatic variant calls. High-throughput single-cell assays also suffered from low mRNA capture efficiency and could not reliably detect lowly expressed genes or resolve transcriptomically similar subclusters.
Tried and failed
differential expression analysis on single-cell RNA sequencing applied to whole tissue disease profiling. Outcome: no signal. Reason: gene dropout effects in single-cell sequencing obscured differences across the whole tissue level
Lost to a baseline
Single-cell RNA sequencing captured ~10% of mRNA, whereas targeted ExSeq yielded ~62% relative to smFISH/ExFISH.
Spatially precise in situ transcriptomics in intact biological systems · MIT
Tried and failed
computational gene genotyping from single-cell RNA sequencing applied to polymorphic immune receptor region. Outcome: data insufficient. Reason: sequencing read coverage mapping to the target region was too shallow
Modelling the impact of iKIRs in adaptive immune responses · Imperial
Tried and failed
single-cell RNA sequencing transcriptomic clustering applied to rare cytokine-expressing immune cell subsets. Outcome: no signal. Reason: mRNA expression dropped below detection limits despite measurable circulating protein levels
The Ecological and Temporal Genetic Architecture of The Microbiome in Health and Disease · Harvard
Tried and failed
variant calling from single-cell RNA sequencing applied to mutational signature detection in lung cancer. Outcome: no signal. Reason: RNA sequencing artifacts generated pervasive noise spectra, obscuring expected somatic mutational signatures
Considered and rejected
Considered and rejected: Rejected performing single-cell variant/lineage calling solely from scRNA-seq or scATAC-seq reads due to pervasive allelic dropout and low read depth
Tried and failed
direct somatic variant calling from single-cell sequencing applied to single-cell RNA and ATAC sequencing. Outcome: no signal. Reason: sparse per-cell coverage and allelic dropout caused over 90% false positives and high miss rates
Tried and failed
single-cell RNA sequencing transcript detection applied to low-abundance functional receptor targets. Outcome: no signal. Reason: Transcript dropout or low mRNA abundance despite confirmed surface protein expression and functional activity
Investigation of maturation and survival of human long-lived plasma cells using integrated single-cell analysis · Georgia Tech
Lost to a baseline
Slide-seq (version 1) had approximately 2.7% the capture efficiency of Drop-seq single-cell RNA sequencing.
Technologies for assaying the spatial position of biomolecules in situ · Harvard
Considered and rejected
Considered and rejected: Smart-seq2 without UMIs for high-throughput single-cell experiments (suffered from low throughput of ~few hundred cells per plate, PCR amplification bias, and inefficient template switching for lowly expressed transcripts)
Considered and rejected
Considered and rejected: Rejected using single-cell RNA-seq to design chemokine pools because it lacks total tissue secretome abundance and misses low-abundance transcripts.
Development of a class B evasin-derived peptide for inhibiting islet inflammation · Oxford
Considered and rejected
Considered and rejected: Rejected degenerate oligonucleotide-primed PCR (DOP-PCR) whole-genome amplification approaches for single-cell sequencing due to poor genomic coverage and allele dropout.
Genomic Instability at Single Cell Resolution · ResearchWorks
Considered and rejected
Considered and rejected: Decided against using single-cell Hi-C to assess single-allele TAD configurations due to low sequencing depth per single cell.
Characterizing the role of condensin II in interphase chromosome conformation and compartmental contacts · ScholarlyCommons at Penn
Tried and failed
droplet-based single-cell RNA sequencing applied to homogeneous cell subpopulation clustering. Outcome: no signal. Reason: insufficient sequencing depth to resolve transcriptomically similar subclusters
Dissecting The Mechanisms of Transitional B Cell Tolerance · Harvard
Quality filtering thresholds and doublet contamination cause clustering artifacts
Rigid quality control and feature filtering inadvertently eliminated valid cell populations with low baseline counts or low dataset variance. At the same time, doublet barcode collisions and gene misannotations created artificial cell clusters, while standard marker co-expression filtering falsely removed legitimate multi-lineage cells.
Lost to a baseline
Single-cell bisulfite sequencing (scBS-seq) libraries were excluded if they showed less than 8% mapping efficiency, less than 100,000 total sequences, greater than 60% methylated CpGs, or average X chromosome CpG island methylation rates greater than 21% (to remove potential somatic granulosa cell contamination).
Molecular profiling of AMA oocytes: from basic biology to improving fertility in aged women · Research Repository UCD
Lost to a baseline
Single-cell transcriptomes with >5% mitochondrial counts, >4% dissociation signature read fraction, or >40,000 unique RNA features excluded during scRNA-seq quality control.
CROSS-COMPARTMENT IMMUNOLOGY OF PEDIATRIC NEUROINFLAMMATORY DISEASE · Penn
Tried and failed
Single-cell RNA sequencing unsupervised clustering applied to Single-cell gene expression profiles. Reason: Misannotated genes artifactually drove false cluster segregation
Investigation of maturation and survival of human long-lived plasma cells using integrated single-cell analysis · Georgia Tech
Tried and failed
Single-cell subclustering without doublet filtering applied to combinatorial indexing single-cell sequencing. Reason: Barcode collisions generated spurious hybrid cell clusters that masqueraded as distinct biological cell types
CHROMATIN ACCESSIBILITY CHANGES DURING ADIPOSE BEIGING AND MELANOCYTE STEM CELL ACTIVATION · Cornell
Tried and failed
co-expression filtering of conflicting lineage markers applied to single-cell transcriptomic doublet detection. Outcome: did not generalise. Reason: legitimate cell types naturally co-expressing multiple tissue programs were falsely classified as doublets and removed
Understanding Gene Regulation In Development And Differentiation Using Single Cell Multi-Omics · Penn
Tried and failed
Uniform quality filtering thresholds across heterogeneous samples applied to single-cell RNA sequencing data. Reason: Discarded valid cell subpopulations with naturally lower transcriptional output and baseline counts
Lost to a baseline
23 cells identified as doublets/sorting artifacts with high read counts and non-specific marker gene expression were removed from single-cell downstream analysis.
Mechanisms of DNA double strand break-mediated neurotoxicity in neurodegenerative disease · MIT
Lost to a baseline
Single-cell gene expression clustering alone failed to identify FOXP3 as canonical for Tregs or distinguish normal from tumor epithelial cells (unlike VIPER inferred protein activity).
Considered and rejected
Considered and rejected: Rejected assuming a uniform cell type distribution when identifying marker genes from unnormalized dispersion metrics in highly imbalanced single-cell datasets.
APPLICATIONS OF SINGLE-CELL GENOMICS IN THE STUDY OF CIRCADIAN AND VASCULAR BIOLOGY · Penn
Tried and failed
quantifying exonic-only reads in single-cell transcriptomics applied to single-cell RNA-seq clustering. Outcome: no signal. Reason: excluding intronic reads lost crucial transitional cell markers, merging distinct sub-populations into canonical clusters
MULTI-MODAL PROFILING OF HUMAN PANCREATIC ISLETS IN NON-DIABETIC AND TYPE 2 DIABETIC STATES · Penn
Tried and failed
high variance feature filtering before dimensionality reduction applied to single-cell RNA sequencing data. Outcome: worse than baseline. Reason: marker genes for rare subpopulations have low overall dataset variance and get excluded
Pseudoreplication and small sample sizes distort differential expression analysis
Performing cell-level differential expression tests without accounting for between-replicate variance led to pseudoreplication and severe p-value inflation. Conversely, aggregating cells into pseudobulk or fractions suffered from low biological replicate counts and high inter-sample variation that obscured true differences.
Tried and failed
pseudobulk differential expression analysis applied to single-cell RNA sequencing data. Outcome: data insufficient. Reason: low recovered cell counts per cell type across participant groups
Single-Cell Biology of Respiratory Viral Infections in the Nasal Mucosa · Harvard
Tried and failed
differential expression with small sample sizes applied to single-cell RNA sequencing data. Outcome: data insufficient. Reason: low biological replicate counts cause severe statistical underpowering and very high false discovery rates
A novel approach to power analysis for differential expression in scRNA-seq data · Imperial
Considered and rejected
Considered and rejected: Rejected single-cell DE methods that aggregate independent univariate comparisons across individual cells for prioritization because they bias toward abundant cell types and drop subtle multivariate signals.
Multimodal transcriptomic atlases of mouse spinal cord injury · EPFL
Considered and rejected
Considered and rejected: Rejected purely cell-fraction based testing because collapsing single-cell counts into patient-level fractions discards cell-level information and yields insufficient statistical power.
Tried and failed
cell-level differential expression testing without pseudobulking applied to single-cell RNA-sequencing data. Reason: pseudoreplication and unmodelled within-sample correlation caused inflated false positive rates under permutation
Single-cell RNA-sequencing in epilepsy; discovery of cell-types, pathways, and drug targets · Imperial
Tried and failed
pseudobulk differential expression with cell-level replication applied to single-cell RNA sequencing perturbation data. Reason: treating individual cells as independent replicates causes extreme p-value inflation due to pseudoreplication
Deciphering Leukaemogenic Mechanisms through System-Scale Analysis of Single-Cell RNA Sequencing Data · Cambridge
Tried and failed
Cell-level differential expression statistical testing applied to Single-cell transcriptomic perturbation analysis. Reason: Failed to account for between-replicate variance, causing widespread false discovery inflation.
Multimodal transcriptomic atlases of mouse spinal cord injury · EPFL
Lost to a baseline
Single-cell differential expression (MAST) was less reproducible compared to pseudobulk edgeR QLRT baseline across comparative datasets
Molecular Underpinnings of Human Brain Evolution and Cognition at Cellular Resolution · DSpace at UTSWMED
Considered and rejected
Considered and rejected: Treating single cells as independent statistical replicates in differential expression tests (standard single-cell DE without sample covariates), rejected due to pseudoreplication and inflated Type I error in favor of pseudo-bulk aggregation.
Single-cell RNA-sequencing in epilepsy; discovery of cell-types, pathways, and drug targets · Imperial
Tried and failed
hierarchical pseudobulk differential abundance analysis applied to single-cell RNA-seq treatment response data. Outcome: did not generalise. Reason: high inter-sample tissue variation produced counterintuitive marker enrichment across response groups
Tissue dissociation and intact cell isolation cause stress artifacts and cell loss
Enzymatic digestion and mechanical sorting induced strong cellular stress responses, biased cell survival, and damaged fragile or large cell types. Dissociating whole cells also stripped away spatial context and prevented the isolation of multinucleated or biobanked frozen tissues.
Considered and rejected
Considered and rejected: Rejected whole-cell single-cell RNA sequencing (scRNAseq) due to massive loss of LEC viability from shear stress during sorting and enzymatic dissociation bias.
Pulmonary lymphatic remodeling in response to influenza-induced inflammation · OpenBU
Tried and failed
dissociated single-cell RNA sequencing applied to spatially localized regional gene expression differences. Outcome: no signal. Reason: loss of spatial context during tissue dissociation obscured sharp anatomical boundary marker expression
SPATIOTEMPORAL PROFILING OF GENE EXPRESSION IN HEALTH AND DISEASE · Cornell
Tried and failed
enzymatic tissue dissociation for single-cell genomics applied to dense extracellular matrix tissue. Reason: enzymatic digestion induced severe cellular stress-response transcriptional artifacts compared to direct nuclear lysis
A Multimodal Molecular View of Human Cartilage Development · Harvard
Considered and rejected
Considered and rejected: Rejected running single-cell RNA-seq on intact cells in favor of single-nuclei RNA-seq (snRNA-seq) due to the large physical size of adult cardiomyocytes (>100um) and reliance on frozen biobanked tissue.
Understanding the genomic and transcriptomic landscape of cardiomyopathy · Imperial
Considered and rejected
Considered and rejected: Single-cell RNA sequencing was rejected for placental syncytiotrophoblasts because multinucleated spread morphology hinders intact single-cell isolation.
Considered and rejected
Considered and rejected: Single-cell enzymatic/mechanical dissociation rejected due to dissociation-induced immediate-early gene artifacts and motoneuron death in favor of single-nucleus isolation.
SPINAL CORD BIOLOGY AT A SINGLE CELL RESOLUTION: INTRINSIC POTENTIAL FOR NEURONAL DEGENERATION AND REGENERATION · JScholarship
Considered and rejected
Considered and rejected: Rejected conventional single-cell RNA sequencing (scRNA-seq) with enzymatic collagenase digestion, because enzymatic digestion introduces dissociation bias and requires immediate processing of fresh viable cells (>90% viability).
Considered and rejected
Considered and rejected: Single-cell RNA sequencing (scRNA-seq) was rejected due to lack of viable enriched live cells from long-term cryopreserved acute PBMC samples.
Trajectory inference and RNA velocity fail to reconstruct developmental dynamics
Dynamical RNA velocity models generated reversed or inverted differentiation vectors due to uninformative splicing dynamics and violated kinetic assumptions. Trajectory reconstruction also failed when intermediate progenitor populations were scarce or when observation models correlated with technical library size.
Tried and failed
dynamical RNA velocity modeling applied to single-cell lineage trajectory inference. Reason: reversed or failed to identify expected biological trajectories across differentiating cellular lineages
Single-Cell Analysis of the Transcriptional and Regulatory Landscape of Cell Differentiation · Cornell
Tried and failed
trajectory inference from single-cell epigenetic profiles applied to adipocyte progenitor differentiation. Outcome: data insufficient. Reason: insufficient intermediate progenitor cell numbers prevented continuous trajectory reconstruction in low-dimensional embeddings
CHROMATIN ACCESSIBILITY CHANGES DURING ADIPOSE BEIGING AND MELANOCYTE STEM CELL ACTIVATION · Cornell
Tried and failed
multimodal single-cell RNA velocity inference applied to differentiating cell state trajectories. Outcome: no signal. Reason: uninformative gene splicing dynamics produced inverted differentiation velocity vectors
Dynamics of the gene regulatory landscape in human spermatogenesis · Imperial
Tried and failed
multimodal data integration for lineage reconstruction applied to single-cell lineage tracing. Outcome: worse than baseline. Reason: incorporating additional transcriptomic data reduced reconstruction accuracy compared to using lineage barcodes alone
Computational Modeling of Spatiotemporal Dynamics in Single-Cell Omics Data · Georgia Tech
Tried and failed
three-state stochastic promoter model with positive feedback applied to longitudinal single-cell transcript count trajectories. Outcome: did not converge. Reason: model architecture could not capture the full time-course trajectory dynamics of transcript counts
Transcriptional regulation of the HTLV-1 provirus · Imperial
Tried and failed
inferring transcription factor regulatory activity from expression applied to single-cell developmental trajectory reconstruction. Outcome: no signal. Reason: Target gene expression enrichment failed to detect activity despite adequate transcription factor transcript abundance
The Transcriptional Biographies of Embryonic Cells · Harvard
Tried and failed
biophysical differential equation dynamical modeling applied to perturbed single-cell periodic processes. Outcome: did not generalise. Reason: underlying biophysical assumptions of transcription kinetics were violated by splicing machinery perturbations
Tried and failed
individual gene binomial observation models applied to single-cell trajectory and tree inference. Outcome: no signal. Reason: failed to capture exposure variations, causing inferred structures to correlate with library size and sparsity instead of biology
Technical batch effects and integration forcing obscure true biological variation
Batch correction and reference mapping algorithms failed to remove technical sample variation without forcing artificial alignments or flattening biological differences. Supervised models and distance metrics also overfit to patient-specific confounders or conflated sequencing depth with biological lineage.
Tried and failed
supervised classifier for differential cell response prioritisation applied to single-cell RNA-sequencing data. Outcome: overfit. Reason: model overfit to patient-specific confounders, assigning high significance even under label permutation
Tried and failed
reference mapping algorithms for cell type annotation applied to single-cell transcriptomic data integration. Outcome: did not generalise. Reason: insufficient resolution without artifactual batch-integration forcing
Considered and rejected
Considered and rejected: Single-cell transcriptomic analyses performed per individual sample independently were rejected because inter-sample biological variation prevented drawing generalisable cross-sample conclusions.
Molecular characterisation of ASXL1-mutant clonal haematopoiesis · Oxford
Considered and rejected
Considered and rejected: Single-cell transcriptomics without feature barcoding, rejected due to technical noise and severe batch effects across multiple knockout populations
Functional impact of inactivating mutations in epigenetic regulators in cancer · Imperial
Tried and failed
graph embedding of sparse feature matrices applied to single-cell chromatin accessibility data. Outcome: did not generalise. Reason: failed to eliminate technical batch effects across samples
Single-Cell Analysis of the Transcriptional and Regulatory Landscape of Cell Differentiation · Cornell
Tried and failed
batch-correction algorithms on multi-source data applied to single-cell RNA-sequencing integration. Outcome: did not generalise. Reason: algorithms failed to eliminate technical batch variance while preserving true biological variation across datasets
Considered and rejected
Considered and rejected: Rejected regressing out cell cycle genes during single-cell preprocessing because it failed to resolve confounding cycle-stage distinctions
Transcriptional regulation of memory B cell heterogeneity · Imperial
Considered and rejected
Considered and rejected: Rejected using raw Euclidean distance to cluster PBMC single-cell profiles, as it conflates sample depth/abundance with lineage distances unlike transcriptome-wide impact
Integrative statistical methods for human biology, from biobanks to perturbation atlases · Harvard
Spatial transcriptomic mapping and cell deconvolution produce inaccurate local estimates
Reference-based transcriptomic deconvolution algorithms yielded inaccurate cell abundance estimates compared to spatial proteomics ground truth. Additionally, spatial correlation and cohort-level averaging obscured localized cellular heterogeneity and produced high density variance.
Tried and failed
reference-based transcriptomic deconvolution algorithms applied to spatial transcriptomics cell abundance estimation. Outcome: did not generalise. Reason: algorithms yielded inaccurate and inconsistent abundance estimates when benchmarked against ground-truth spatial proteomics single-cell data
Tried and failed
reaction-diffusion simulation from single-cell expression profiles applied to spatial cell density and neighborhood reconstruction. Outcome: data insufficient. Reason: missing key chemokine-producing cell types in single-cell data led to incorrect spatial co-localization
Bridging single-cell genomics and tissue spatial organization in health and cancer · Harvard
Tried and failed
spatial subsampling for abundance estimation applied to spatially correlated single-cell tissue data. Outcome: data insufficient. Reason: spatial correlation drastically reduced effective sample size, causing high variance in density estimates
Geometry of single-cell multiplex data reflects biophysical processes in cells · Harvard
Tried and failed
spatial correlation search in reconstructed transcriptomes applied to single-cell RNA sequencing data. Outcome: no signal. Reason: spatial correlation queries failed to identify co-patterned genes along the inferred axis
Tried and failed
cohort-level spatial neighborhood composition analysis applied to single-cell spatial transcriptomics data. Outcome: no signal. Reason: Averaging spatial neighborhoods across entire cohorts obscures localized cellular microenvironment heterogeneity between distinct transcriptional programs
Left open by the authors
Problems the authors named and did not get to.
Left open
Re-analyze the scRNA-seq dataset by relaxing mitochondrial DNA filter thresholds to recover depowered neuronal populations. Blocker: Requires access to the private single-cell RNA sequencing dataset generated in the thesis
Left open
Perform single-cell RNA sequencing to map genetically distinct substantia nigra reticulata neuronal populations to functional and projection phenotypes. Blocker: Requires wet lab facilities, tissue preparation, and single-cell RNA sequencing apparatus/reagents.
Substantia nigral activity in self-timed movements · Harvard
Left open
Perform single-cell heritability enrichment profiling on the nucleus basalis of Meynert and locus coeruleus to evaluate cell-type vulnerability in neurodegeneration. Blocker: Requires postmortem human brain tissue access and wet-lab single-nucleus RNA sequencing of specific brain regions
Cell states and neuronal vulnerabilities in neurodegenerative diseases · Harvard
Left open
Compare single-cell RNA-seq signatures of iPSC-derived brain cells directly with single-nucleus RNA-seq datasets from human postmortem Alzheimer's brains. Blocker: Human postmortem snRNA-seq data (e.g., ROSMAP or AMP-AD) and specific iPSC scRNA-seq datasets typically require controlled-access data use agreements (DUAs).
Left open
Perform single-cell and bulk RNA sequencing of spike- and non-spike-specific T cells to profile transcriptional mechanisms of synergy. Blocker: Requires wet-lab execution of single-cell/bulk RNA sequencing on biological samples
Rational Design of CD8+ T Cell Vaccines for SARS-CoV-2 · Harvard
Left open
Characterize immune cell phenotypes using CyTOF or single-cell multi-omics methods like CITE-seq and SEC-seq. Blocker: Requires wet lab experimental facilities, biological samples, and specialized instrumentation (e.g., CyTOF/Hyperion Imaging System, single-cell sequencing platforms).
Utilizing Combinatory Adjuvant-Loaded Chitosan-Derived Nanoparticles for a Joint SARS-CoV-2/Influenza Vaccine · Georgia Tech
Left open
Disentangle lineage relationships between early effector and chronic exhausted CD8+ T cell subsets using scRNA-seq trajectory analysis, spatial transcriptomics, or CRISPR barcoding. Blocker: Requires wet lab experimentation involving single-cell spatial transcriptomics, single-cell RNA-seq, or CRISPR barcoding in infection models.
Cell Intrinsic Factors Influence CD8+ T Cell Fate Decisions Following Acute and Chronic Infection · Cornell
Left open
Perform phenotyping and single-cell TCR sequencing on tumor-infiltrating lymphocytes induced across different mimotope vaccine regimens to assess expansion and exhaustion. Blocker: Requires a wet lab, animal models, vaccine administration, cell isolation, flow cytometry/phenotyping apparatus, and single-cell sequencing equipment.
Left open
Investigate the specific immune cellular phenotypes driving memory response and tumor rejection following anti-PD-1 therapy re-challenge. Blocker: Requires a wet lab, animal models, and cellular immunology assays (e.g., flow cytometry or single-cell sequencing).
An Affinity Threshold for Maximum Efficacy in Anti-PD-1 Cancer Immunotherapy · MIT
Left open
Quantify phosphosignaling and cytokine expression changes at single-cell resolution after repetitive mild traumatic brain injury using flow cytometry or single-cell sequencing. Blocker: Requires performing wet lab experiments (flow cytometry or single-cell RNAseq/proteomics) on mouse brain tissue.
Profiling the neuroimmune cascade after repetitive mild traumatic brain injury · Georgia Tech
Checking a claim in this area?
We can run the same search on any method or claim. If nothing turns up, we will say so, and that proves nothing on its own.