Built by Mike Kazemi (mike.kazemi@mail.utoronto.ca) · Morris Lab · University of Toronto
Overview
BloodNNet integrates genome-wide fine-mapping of 29 blood cell traits with deep learning–based chromatin effect scores, enhancer-to-gene (E2G) predictions, and multi-omic epigenomic annotations to help researchers prioritize regulatory variants for functional follow-up. For each prioritized variant, the tool designs CRISPRi guide RNAs (gRNAs) targeting the enhancer containing the variant in K562 cells, and generates publication-quality locus visualizations. Researchers can filter variants through an interactive dashboard using either a SNP-centric or enhancer-centric approach, then download gRNAs and locus plots for their selected candidates.
GWAS Fine-mapping Data
Genetic variants and their fine-mapping statistics come from the large-scale blood cell trait GWAS by Vuckovic et al. (2020) [1], which studied 29 blood cell phenotypes across 563,085 European-ancestry participants from the UK Biobank and international collaborators. Fine-mapping was performed using FINEMAP v1.3.1. BloodNNet includes all variants with a posterior inclusion probability (PP) ≥ 0.001 in at least one of the 29 traits. The maxPIP filter in the dashboard reflects the highest PP for a given variant across all traits it is associated with.
Trait Abbreviations
| Trait | Abbreviation |
|---|---|
| Platelet | |
| Platelet count | plt |
| Mean platelet volume | mpv |
| Platelet distribution width | pdw |
| Plateletcrit | pct |
| Mature red cell | |
| Red blood cell count | rbc |
| Mean corpuscular volume | mcv |
| Hematocrit | hct |
| Mean corpuscular hemoglobin | mch |
| Mean sphered corpuscular volume | mscv |
| Mean corpuscular hemoglobin concentration | mchc |
| Hemoglobin concentration | hgb |
| Red cell distribution width | rdw_cv |
| Immature red cell | |
| Reticulocyte count | ret |
| Reticulocyte fraction of red cells | ret_p |
| Immature fraction of reticulocytes | irf |
| High light scatter reticulocyte count | hlr |
| High light scatter reticulocyte % of red cells | hlr_p |
| Mean reticulocyte volume | mrv |
| Myeloid white cell | |
| Monocyte count | mono |
| Neutrophil count | neut |
| Eosinophil count | eo |
| Basophil count | baso |
| Lymphoid white cell | |
| Lymphocyte count | lymph |
| Compound white cell | |
| White blood cell count | wbc |
| Monocyte % of white cells | mono_p |
| Neutrophil % of white cells | neut_p |
| Eosinophil % of white cells | eo_p |
| Basophil % of white cells | baso_p |
| Lymphocyte % of white cells | lymph_p |
ChromBPNet Chromatin Effect Scores
ChromBPNet [2] is a deep learning model that predicts base-resolution chromatin accessibility profiles from DNA sequence. For each variant, ChromBPNet scores the difference in predicted chromatin accessibility between the reference and alternative alleles, providing an in silico measure of regulatory impact. Three complementary score types are reported per cell type:
log_counts_diff — log-fold difference in predicted total accessibility signal.
log_probs_diff_abs_sum — summed absolute difference in predicted nucleotide-resolution accessibility profile.
probs_jsd_diff — Jensen-Shannon divergence between reference and alternative accessibility profiles.
Models were trained in-house on ATAC-seq data from 17 blood and hematopoietic cell types (hg38): B cells (Bcell), CD34+ cells from bone marrow (CD34BM), CD34+ cells from cord blood (CD34CB), CD4 T cells (CD4T), CD8 T cells (CD8T), common lymphoid progenitors (CLP), common myeloid progenitors (CMP), erythroid cells (ERY), granulocyte-monocyte progenitors (GMP), hematopoietic stem cells (HSC), K562, lymphoid-primed multipotent progenitors (LMPP), megakaryocytes (MEGA), megakaryocyte–erythroid progenitors (MEP), monocytes (MONO), multipotent progenitors (MPP), and natural killer cells (NK). The n cells enriched filters show how many of these 17 cell types have a score in the top 10% of all fine-mapped Vuckovic variants, serving as a measure of cell-type specificity.
Enhancer-to-Gene (E2G) Predictions
Enhancer-to-gene regulatory interactions in K562 cells are predicted by ENCODE-rE2G, a supervised machine learning model developed by Gschwind et al. [3] that integrates chromatin state, 3D contact data, and large-scale CRISPR perturbation measurements. Scores range from 0 to 1, where higher values indicate a more confident enhancer–gene regulatory interaction. The genome-wide K562 prediction file (ENCFF950FTI) was downloaded from ENCODE. For each enhancer containing a variant, the top three predicted target genes and their scores are displayed.
Epigenomic Signal Tracks (K562)
Chromatin accessibility measured by ATAC-seq in K562 cells. Data were generated by Bond et al. (2023) [4] in the Phanstiel Lab. Variants overlapping an ATAC-seq peak reside in open chromatin and are more likely to be regulatory.
DNase I hypersensitivity signal in K562 cells from the ENCODE Consortium [5], marking regions of open chromatin accessible to DNase I cleavage.
Fold-change over control ChIP-seq signal for the histone modification H3K27ac in K562 cells (ENCODE [5]). H3K27ac marks active enhancers and promoters.
CTCF ChIP-seq signal in K562 cells (ENCODE [5]). CTCF binding marks topological domain boundaries and insulator elements that constrain enhancer–promoter communication.
MPAC — Allelic Effect Predictions
The MPAC K562 Skew score (k562_skew_pred) is a machine-learning prediction of the allelic effect of a noncoding variant on gene expression in K562 cells. It is defined as the predicted log₂ fold difference in reporter gene expression between the alternative and reference alleles — an in silico measurement of allelic skew. The model was trained on massively parallel reporter assay (MPRA) data and described by Gosai et al. (2024) [6]. Cell-type-specific predictions including K562 are available at zenodo.org/records/15186315. Variants not covered by the MPRA training set carry no prediction (shown as NOT_FOUND or ALLELE_MISMATCH) and can be optionally included or excluded using the checkbox in the MPAC filter panel.
Gene Category Annotations
Protein-coding genes in the locus visualization are color-coded by membership in one or more of four functionally relevant categories. A gene may belong to multiple categories simultaneously, in which case its body is divided into equal-width colored segments.
A curated set of 704 high-confidence causal genes for blood cell traits, assembled from loss-of-function burden tests across 29 traits in the UK Biobank and genes linked to Mendelian blood disorders. Source: Ghatan et al. [7].
Genes with a significant expression quantitative trait locus (eQTL) in blood-related tissues from the eQTL Catalogue, a standardized resource of uniformly processed eQTL datasets across human tissues and cell types.
Genes linked to a regulatory element through a CRISPRi perturbation screen in K562 cells, manually curated from published literature [3,7,8,9,10,11].
Genes identified as essential for cell fitness through Project Achilles (DepMap), which uses genome-scale CRISPR-Cas9 perturbations across hundreds of cancer cell lines to systematically catalog gene essentiality.
CRISPRi gRNA Design
Guide RNAs are designed to silence the enhancer containing each prioritized variant using catalytically dead Cas9 (dCas9) fused to a KRAB repressor domain. The following design parameters are used:
Target window: ±150 bp around the midpoint of the enhancer containing the variant. All SpCas9 NGG-PAM guide sequences within this window on both strands are identified.
CRISPRi blast radius: ±1 kb around the guide binding site. Based on empirical measurements from CRISPRi screens in K562 cells showing that a bound dCas9-KRAB complex silences gene expression within approximately 1 kb on either side [8]. The silenced enhancers column in downloads lists any other enhancers that fall within this radius.
Seed score: The number of GG dinucleotides in the 12 bp PAM-proximal seed region of the spacer. Guides with higher seed scores are more likely to have off-target activity. Based on optimized library design parameters from Srikanth et al. [12]. A maximum seed score of 1 is recommended; score ≥ 2 indicates a likely promiscuous guide.
The FlashFrySEQ column in downloaded gRNA tables provides the full context sequence (spacer flanked by constant sequences) for off-target scoring with FlashFry or Doench 2014 scoring.
References
- Vuckovic, D. et al. The polygenic and monogenic basis of blood traits and diseases. Cell 182, 1214–1231.e11 (2020). doi:10.1016/j.cell.2020.08.008
- Pampari, A. et al. ChromBPNet: bias factorized, base-resolution deep learning models of chromatin accessibility reveal cis-regulatory sequence syntax, transcription factor footprints and regulatory variants. Preprint at bioRxiv (2024).
- Gschwind, A. R. et al. An encyclopedia of enhancer-gene regulatory interactions in the human genome. Preprint at bioRxiv (2023).
- Bond, M. L. et al. Chromatin loop dynamics during cellular differentiation are associated with changes to both anchor and internal regulatory features. Genome Research 33, 1258–1268 (2023). doi:10.1101/gr.277397.122
- ENCODE Project Consortium. An integrated encyclopedia of DNA elements in the human genome. Nature 489, 57–74 (2012). doi:10.1038/nature11247
- Gosai, S. J. et al. Machine-guided design of cell-type-targeting cis-regulatory elements. Nature 634, 1211–1220 (2024). doi:10.1038/s41586-024-08070-z
- Ghatan, S. et al. CRISPRi perturbation screens and eQTLs provide complementary and distinct insights into GWAS target genes. Preprint at bioRxiv (2025).
- Morris, J. A. et al. Discovery of target genes and pathways at GWAS loci by pooled single-cell CRISPR screens. Science 380, eadh7699 (2023). doi:10.1126/science.adh7699
- ter Weele, M. et al. Genome-wide annotation of gene regulatory elements linked to cell fitness. Preprint at bioRxiv (2025).
- Reilly, S. K. et al. Direct characterization of cis-regulatory elements and functional dissection of complex genetic associations using HCR–FlowFISH. Nature Genetics 53, 1166–1176 (2021).
- Yao, D. et al. Multicenter integrated analysis of noncoding CRISPRi screens. Nature Methods 21, 723–734 (2024).
- Srikanth, J. et al. Optimized parameters for Cas9 CRISPR interference library design. Preprint at bioRxiv (2026).