Chapter Four · failure evidence
What Epigenetic & Chromatin Profiling got wrong, from 54 dissertations
Epigenetic and chromatin profiling studies encounter frequent methodological obstacles across assay preparation, computational feature extraction, and predictive modeling. Researchers regularly report that chromatin accessibility, histone modifications, and DNA methylation patterns decouple from downstream gene expression and fail to generalize across independent cohorts. These records come from PhD theses at 16 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.
DNA methylation assays and biomarker models suffer from technical biases and poor generalizability
Conventional cloning, hexamer primer ends, and copy number variations introduce substantial measurement biases into DNA methylation estimates. Biomarker classifiers, clocks, and association models often lose significance after multiple testing correction, underpredict chronological age, or fail to generalize across cohorts.
Tried and failed
conventional bisulfite PCR cloning and sequencing applied to DNA methylation baseline quantification. Reason: cloning and colony-picking bias produced inaccurate methylation estimates
Exploring the function of NLRP7 in regulating human maternal imprinting · Imperial
Tried and failed
gene ontology enrichment on selected regression features applied to blood DNA methylation epigenetic biomarkers. Outcome: no signal. Reason: enrichment terms lost statistical significance after multiple testing correction
Tried and failed
subgroup analysis by single measurement technology applied to meta-analysis of DNA methylation biomarkers. Reason: restricting to a single assay technique did not reduce inter-study effect size heterogeneity
Tried and failed
epigenome-wide association study on longitudinal methylation changes applied to longitudinal whole-blood DNA methylation data. Outcome: no signal. Reason: No individual CpG sites reached genome-wide significance after multiple testing correction.
Tried and failed
regularized linear regression on high-dimensional epigenetic data applied to predicting brain structural MRI volume. Outcome: did not generalise. Reason: blood epigenetic signatures did not capture brain regional neurodegeneration features across cohorts
Tried and failed
single epigenetic marker for disease classification applied to cancer screening validation cohort. Outcome: did not generalise. Reason: marker achieved chance-level performance (AUC = 0.50) in an independent validation cohort
Lost to a baseline
Raw methylation data UMAP embeddings exhibited less defined tissue-specific clusters, stronger batch-specific clustering, and weaker sex-specific separation compared to contextualized MethylGPT sample embeddings.
On the Quantification of Aging · Harvard
Lost to a baseline
The published commercial Infinium Mouse Methylation BeadChip multi-tissue clock (Zhou et al. 2022) uniformly underpredicted chronological age in blood samples
Improved methods for age, frailty, and mortality prediction in large-scale longevity studies · Harvard
Considered and rejected
Considered and rejected: Rejected selecting differentially methylated probes at a global cross-tissue level in favor of tissue-specific differential methylation analysis due to strong tissue-of-origin signals.
SUPERVISED METHYLATION SIGNATURES FOR TISSUE-SPECIFIC CANCER ETIOLOGICAL FACTORS · JScholarship
Considered and rejected
Considered and rejected: Rejected relying on uncorrected read ends due to hexamer primer-induced methylation bias at the 5' and 3' ends.
Evaluation of Computational and Experimental Parameters in RNA Bisulfite Sequencing Analysis and Applications in Brain Development Studies · Virginia Tech
Considered and rejected
Considered and rejected: Rejected strict binary methylation classification (0 or 1) on treated samples because it masks partial (~20%) demethylation changes across the genome.
Elucidating the effects of DNA modification on transcription · Oxford
Considered and rejected
Considered and rejected: Rejected using weighted methylation for GWAS across accessions because it is distorted by structural/copy number variants, using unweighted mean methylation instead.
Genome - epigenome dynamics and natural variation in Thlaspi arvense · Publikationssystem UB Tuebingen
Chromatin conformation mapping and enhancer identification face resolution limits and regulatory discordance
Predicted chromatin loops and candidate enhancer elements frequently lack functional regulatory activity or active histone marks in validation assays. In addition, sequencing-based conformation methods struggle with cost-prohibitive resolution scaling for small loops and loss of fine microcompartment detection with longer fragments.
Tried and failed
epigenomic profiling of regulatory elements applied to rare disease variant enrichment analysis. Outcome: no signal. Reason: chromatin accessibility and histone marks alone insufficiently capture functional disruption by rare non-coding variants
DECIPHERING MULTI-LAYER FUNCTIONAL EFFECTS OF GENOMIC VARIANTS IN HUMAN DISEASES · Cornell
Tried and failed
HiChIP-based identification of candidate functional enhancers applied to developmental cis-regulatory element validation. Outcome: no signal. Reason: Predicted chromatin-interacting regions were inactive or lacked specific regulatory activity in vivo
Decoding the cis-regulatory principles of signaling-responsive neural crest enhancers · Cornell
Tried and failed
biological pathway enrichment on differential chromatin peaks applied to developmental stage-specific open chromatin regions. Outcome: no signal. Reason: small candidate peak sets lacked functional enrichment and appeared to be background noise
Modeling and integration of multi-omics data to study regulatory landscapes governing placenta development · Iowa State
Tried and failed
epigenetic histone mark enrichment as enhancer filter applied to identifying functional regulatory transposable elements. Outcome: no signal. Reason: active elements lacked the expected histone mark, and unmarked elements were statistically more significant
Tried and failed
long-fragment sequencing read analysis applied to chromatin microcompartment resolution. Outcome: no signal. Reason: fragments over 120 bp lost fine resolution and caused microcompartment signals to fade after cohesin depletion
Statistical methods for transcription factor footprinting in 3D genome assays · Harvard
Considered and rejected
Considered and rejected: Rejected using publicly available Hi-C data to annotate super-enhancer promoter loops because enhancer regulation and chromatin looping dynamics are highly cell-type and chromatin-modifier specific.
The Epigenetic Role of Nuclear Receptor Co-Repressor 2 (NCOR2) in Neuroendocrine Prostate Cancer Progression · DSpace at SUNY Buffalo
Considered and rejected
Considered and rejected: Rejected using sequencing (Hi-C/high-throughput 3C-seq) to read out chromatin conformation because map resolution scaling made detecting small ~15kb regulatory loops cost-prohibitive compared to targeted PCR 3C.
Regulation of neural stem cell behaviors in Drosophila · Harvard
Considered and rejected
Considered and rejected: Rejected DNA-FISH measurements across the gene body for chromatin decompaction due to confounding RNAPII-mediated transcription loop extrusion.
Understanding Cell State Transitions in Development and Disease · ResearchWorks
Chromatin accessibility and activity scores fail to predict transcriptional output or gene expression
Gene activity modeling and distance-weighted accessibility aggregations around gene bodies consistently correlate poorly with single-cell RNA levels. Open chromatin peaks and linear proximity to histone acetylation do not reliably reflect transcriptional activation, and promoter accessibility can remain completely unchanged following transcriptional silencing.
Tried and failed
motif-altering regulatory variant mapping to downstream expression applied to transcription factor binding regulatory loci. Outcome: no signal. Reason: chromatin accessibility and histone alterations frequently did not translate into detectable gene expression QTLs
Tried and failed
chromatin accessibility gene score modeling applied to predicting single-cell gene expression. Outcome: worse than baseline. Reason: distance-weighted accessibility aggregation around gene bodies produced negative out-of-sample R2 scores
Integrating genomic and multiomic data for computational analysis of gene regulation in circulating immune cells · Georgia Tech
Tried and failed
Gene activity scoring from chromatin accessibility applied to predicting single-cell gene expression. Outcome: no signal. Reason: Accessibility scores correlated poorly with measured RNA levels across multiple scoring algorithms
Integration of single-cell multimodal data to define the chromatin landscape of rheumatoid arthritis · Harvard
Tried and failed
exponential distance-decay gene activity modeling applied to predicting single-cell gene expression from chromatin accessibility. Outcome: no signal. Reason: modeled accessibility scores correlated poorly with measured mRNA expression levels
Single-Cell Genomics of Human Cancer · Harvard
Tried and failed
open chromatin peak based enhancer selection applied to cell type specific viral promoter design. Outcome: no signal. Reason: high chromatin accessibility did not correlate with transcriptional activation strength
Tried and failed
nearest-neighbor feature assignment on 1D genomic sequence applied to predicting gene expression from enhancer acetylation. Outcome: no signal. Reason: linear proximity fails to capture functional distal 3D chromatin interactions regulating target promoters
Tried and failed
using chromatin accessibility to measure promoter inactivation applied to enhancer knockout transcriptional silencing. Outcome: no signal. Reason: promoter accessibility remained unchanged despite near-complete loss of transcription and active histone marks
Sample preparation and experimental processing protocols compromise chromatin profiling assays
Excessive cross-linked cell density impedes sonication shearing to abolish enrichment, while chopped tissue preparations and low nuclear recovery degrade peak signals and amplify batch effects. Standard library-size normalization severely underestimates binding during global factor loss, and low enzyme concentrations yield inadequate cleavage resolution.
Tried and failed
chromatin immunoprecipitation sequencing applied to transcription factor profiling. Outcome: no signal. Reason: samples failed standard quality control metrics
Transcription factor antagonism regulates heterogeneity in embryonic stem cell states · MIT
Considered and rejected
Considered and rejected: Rejected standard library-size normalization alone for mitotic ChIP-seq because global reduction in chromatin-bound factor causes severe underestimation (MAPD 123-145% vs 62-68% for qPCR scaling).
Mining The Spatial And Temporal Context By Which Transcription Factors Occupy Chromatin · Penn
Tried and failed
enzymatic cleavage assay at low concentrations applied to single-locus chromatin accessibility quantification. Outcome: no signal. Reason: low cleavage probability at discriminative enzyme concentrations led to poor measurement resolution
Principled Methods and Models for Deep Learning Based Functional Genomics · MIT
Tried and failed
Sonication of high cell density cross-linked chromatin applied to chromatin immunoprecipitation fragmentation. Outcome: no signal. Reason: Excessive cell concentration prevented efficient shearing, leaving fragments over 1 kb and abolishing qPCR enrichment
Considered and rejected
Considered and rejected: Standard ChIP-seq sonication fragment size (200-300 bp) was avoided/rejected initially because it is too short to retain unique mappable flank sequences outside extensive tandem repeats; chromatin was sonicated to 0.5-2 kb first.
Defining Sites Of Replication Fork Collapse Caused By Atr Inhibition · Penn
Tried and failed
chopped tissue sample preparation for ATAC-seq applied to plant seedling chromatin accessibility profiling. Outcome: no signal. Reason: unusually low signal portion of tags score during peak calling
Plastid to Nucleus Signalling and the Evolution of C4 Photosynthesis · Cambridge
Tried and failed
subclustering single-nucleus chromatin accessibility data applied to infiltrating immune cell subpopulations. Outcome: data insufficient. Reason: low single-nucleus recovery caused batch effects to overpower biological cell-type separation
Computational footprinting and motif enrichment fail to reliably identify physical factor binding
Computational transposase footprinting scores fail to correlate with physical protein binding dynamics and lose detection signal after adjusting for sequencing depth. Unfiltered motif scans yield excessively broad lists, while motif accessibility and transcription factor expression poorly align with actual genomic occupancy or spatial interactions.
Tried and failed
lineage-defining transcription factor binding analysis applied to differential chromatin accessibility identification. Outcome: no signal. Reason: The target factor lacked pioneer activity and binding did not distinguish differentially accessible chromatin sites
Decoding Regulatory T Cell Transcription Factor Networks: From Identity to Diversity · Harvard
Tried and failed
correlating transposase footprinting activity with proximity-labeled protein abundance applied to measuring transcription factor chromatin binding dynamics. Outcome: no signal. Reason: computational footprint scores did not correlate with physical chromatin-associated transcription factor abundance measured by mass spectrometry
Accelerating the functional prioritization of gene regulatory biomarkers in acute myeloid leukemia · Harvard
Tried and failed
Unfiltered transcription factor motif enrichment analysis applied to accessible chromatin regions in single-cell data. Reason: Screening all database motifs yielded overly broad, non-specific candidate lists without rank-based filtering
Exaptation of ancestral cell-identity networks enables C4 photosynthesis. · Cambridge
Tried and failed
motif-centered spatial interaction enrichment analysis applied to candidate chromatin architectural transcription factors. Outcome: no signal. Reason: candidate factor motifs showed no enrichment for long-range interactions across tested conditions
Statistical methods for transcription factor footprinting in 3D genome assays · Harvard
Tried and failed
chromatin accessibility footprinting algorithms applied to transcription factor binding site detection. Outcome: no signal. Reason: controlling for sequencing cut coverage depth eliminates most footprinting signals
Characterizing Regulatory Elements and Non-Coding Variants in the Human Genome · Harvard
Considered and rejected
Considered and rejected: Rejected simple correlation between TF expression and motif accessibility in open chromatin; required per-locus linear regression clustering because TF expression did not reliably align with motif accessibility alone.
A Multimodal Molecular View of Human Cartilage Development · Harvard
Fixed binning and peak calling algorithms fail to capture complex epigenetic signal distributions
Uniform binning and narrow-peak callers miss variable signal boundaries and fail to model diffuse, broad heterochromatic domains. Conversely, excessively large windows reduce resolution without performance gains, leading researchers to favor continuous raw alignment profiles over discrete peak calls.
Tried and failed
uniform fixed-size binning instead of peak detection applied to chromatin accessibility feature extraction. Outcome: worse than baseline. Reason: Fixed windows fail to capture variable-width signal boundaries, missing more than half of active regions.
Tried and failed
narrow-peak calling on broad epigenetic enrichment domains applied to broad histone modification profiling. Outcome: no signal. Reason: algorithm model assumes narrow enriched regions rather than diffuse, broad heterochromatic enrichment domains
Decoding epigenetic adaptation to adjuvant endocrine therapies in hormone-dependent breast cancer · Imperial
Considered and rejected
Considered and rejected: Window sizes greater than 600 bp (tested 1000 bp) for chromatin state prediction were rejected due to diminishing performance returns on validation sets while decreasing effective resolution.
From Arabidopsis to Zea: Learning conserved cis mechanisms of gene regulation · Cornell
Tried and failed
narrow peak calling algorithms applied to broad genomic enrichment domains. Outcome: no signal. Reason: narrow peak callers fail to accurately capture diffuse, broad epigenetic deposition patterns
Considered and rejected
Considered and rejected: Peak calling analysis (MACS2 broad peaks) was rejected in favor of full continuous raw alignment profile analysis because epigenetic landscapes act as continuous analog noise histograms rather than discrete peaks.
Differential DNA methylation changes show poor correlation with differential gene expression
Differentially methylated regions and parent-of-origin methylation marks exhibit negligible direct overlap with differentially expressed genes in host response and cross-breeding studies. Furthermore, DNA methylation fails as a proxy for transcriptional activity because it does not directly capture transcription from inactive chromosomes.
Tried and failed
correlating differential DNA methylation with differential expression applied to host response to bacterial infection. Outcome: no signal. Reason: differential methylation changes showed no direct overlap with differentially expressed genes
Functional Epigenomics In Insects Using Next-Generation Sequencing Methods · Georgia Tech
Tried and failed
correlating differential DNA methylation with differential gene expression applied to identifying functionally regulated candidate genes. Outcome: no signal. Reason: very small direct overlap between differentially methylated region genes and differentially expressed genes
Hippocampal Epigenomic and Transcriptional Plasticity of Cocaine Seeking and Extinction · Cornell
Tried and failed
correlating parent-of-origin differential methylation with gene expression applied to reciprocal cross epigenetic gene regulation. Outcome: no signal. Reason: differential methylation did not correlate with parent-of-origin differentially expressed genes
Functional Epigenomics In Insects Using Next-Generation Sequencing Methods · Georgia Tech
Considered and rejected
Considered and rejected: Rejected DNA methylation data for compiling XCI status annotations because methylation does not directly measure transcription from Xi
Weakening of Y promoters shaped the evolution of human sex chromosomes · MIT
Left open by the authors
Problems the authors named and did not get to.
Left open
Probe whether replisomes push cohesin over 20 kb to termination sites using ChIP assays at short intervals without transcription. Blocker: Requires wet lab facilities and yeast strains to perform chromatin immunoprecipitation (ChIP) assays
Replisome components in sister chromatid cohesion establishment · Imperial
Left open
Identify specific target genes epigenetically regulated by histone lactylation in Plasmodium falciparum using genomic profiling methods like ChIP-seq. Blocker: Requires wet-lab parasite cell culture, chromatin immunoprecipitation (ChIP-seq/CUT&RUN), and sequencing for histone lactylation marks
Histone Lactylation Dynamics in Plasmodium falciparum · Harvard
Left open
Profile Cited2-chromatin interactions using CUT&RUN or ChIP-seq and measure histone acetylation via CUT&Tag in DRG neurons. Blocker: Requires wet-lab experimental facilities, mouse models, and molecular biology assays (CUT&RUN, ChIP-seq, CUT&Tag)
The role of Cited2 in axonal regeneration · Imperial
Left open
Determine other histone modifications involved in seed germination and characterize combinatorial epigenetic regulation of gene expression. Blocker: Requires molecular biology wet-lab experiments (e.g., ChIP-seq for various histone marks, plant assays in Arabidopsis).
Functional characterization of histone binding protein ENAP1 in seed germination and drought response · UT Austin
Left open
Perform ChIP-seq on mouse tissues to directly determine chromatin and histone modifications induced by combined oral contraceptive pill treatment. Blocker: Requires wet-lab facilities, animal tissue samples, and wet-lab ChIP-seq experimental protocols
Molecular mechanisms of ovarian cancer prevention · Imperial
Left open
Replicate ChIP-seq experiments across 1 to 7 hours post-infection to confirm CTCF binding kinetics in HSV-1 infected chromatin. Blocker: Requires wet lab facilities, HSV-1 viral cultures, and chromatin immunoprecipitation (ChIP-seq) experiments
Left open
Add read de-duplication and chromatin accessibility clustering to a scATAC-seq pipeline to separate malignant from non-malignant cells across tumor datasets. Blocker: None
Mutations and disorder in cellular systems · Imperial
Left open
Quantify the direct causal effects of structural variants on DNA methylation patterns and target gene expression across cancer patient cohorts. Blocker: None
Left open
Profile genome-wide DNA methylation across the matched longitudinal patient cohort using EM-Seq or whole-genome bisulfite sequencing. Blocker: Requires access to physical patient tissue samples and a wet lab for DNA methylation sequencing
Identification and characterisation of somatic regulatory mutations in the breast cancer genome · Imperial
Left open
Validate multi-omics survival-associated clusters on external patient cohorts using miRNA, mRNA, and DNA methylation data. Blocker: Specific external cohorts with matching multi-omics modalities and clinical outcomes are not identified
Informatics Approaches for Identifying Drug-Specific Markers and Deciphering Genetic Regulation Mechanism in Cancer Treatment · 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.