DNA Repair Mechanisms Analysis
Single-cell analysis for a CRISPRi study of staggered- versus blunt-end DNA repair (a lab colleague's PhD project, Chakraborty lab, CSIR-IGIB)
Project Overview
This page describes the computational analysis I did for a PhD study in Dr. Debojyoti Chakraborty's lab at CSIR-IGIB. The study asked which DNA-repair genes matter for accurately repairing a Cas9-made double-strand break, and whether the answer differs between staggered (overhanging) and blunt DNA ends. Knowing this could help make genome editing more precise.
A lab colleague led the study as her PhD project (Chapter 3 of her thesis): she carried out the experiments and generated the single-cell RNA-seq data. Her HEK293T cells expressed a catalytically dead Cas9 repressor (dCas9-KRAB) and carried a pooled CRISPRi library of guides against DNA-repair genes. She transfected them with a fluorescent ZsGreen reporter plasmid that had been cut beforehand to leave staggered ends (with en1FnCas9, an engineered FnCas9 variant) or blunt ends (with SpCas9). She then sorted the cells that had accurately repaired the reporter from those that had not, and profiled them by BD Rhapsody single-cell RNA-seq. An earlier single-cell analysis of the data is also acknowledged in the thesis.
From October 2024 to January 2025 I re-analysed the data, starting from the BD Rhapsody count matrices. I did the quality control, integration and differential expression in Seurat, compared gene sets between the staggered- and blunt-end conditions, and ran Gene Ontology enrichment. I also made the heat map and volcano plots in her thesis (Fig. 3.8, credited to me).
The analysis compares sorted populations, not individual cells. Each condition was a single sample, cells were not linked to their CRISPRi guide, and the candidate genes it points to still need experimental validation.
Re-analysis from Count Matrices
Seurat quality control, integration and differential expression of each sorted population against the control sample
Staggered versus Blunt Ends
Gene-set comparisons between cells that did and did not accurately repair the reporter, for each end type
Enrichment and Figures
Gene Ontology enrichment (DAVID) and the heat map and volcano plots used in the thesis
Separately, in May–June 2024, I did an analysis for another chapter of the same thesis. I screened ClinVar's pathogenic 1-, 2- and 3-bp deletions (the May-2024 release, on GRCh38) for an NGG PAM placed so that a Cas9 cut followed by templated fill-in could, in principle, re-add the lost bases, and made the plots for it, re-plotted for the thesis in January 2025 (Fig. 5.2 of her thesis, credited to me). It is a census of where a suitable PAM exists on the reference plus strand; it does not model the repair product or check the protein.
Computational Pipeline
The single-cell processing was a scripted Seurat 4 pipeline on the institute's HPC cluster. I did the gene-set comparisons and figures in Python and the enrichment in DAVID.
Quality Control and Integration
Per-sample cell filters (500–6,500 genes, 1,000–45,000 UMIs and under 12% mitochondrial reads per cell), log-normalisation, and Seurat's standard CCA-anchor integration of the five samples, with conservative settings.
Differential Expression
Seurat FindMarkers (Wilcoxon rank-sum test) comparing each sorted population with the control sample.
Gene-Set Comparison
Among the CRISPRi library genes, I took those expressed at lower levels than in the control in cells that had not repaired the reporter, checked their status in the cells that had repaired it, and compared the resulting lists between staggered and blunt ends.
Enrichment and Figures
Gene Ontology over-representation analysis (DAVID) of the resulting lists, and the heat map, volcano plots and Venn diagrams, made in Python.
Limits: each condition was a single sample, so cells served as replicates; differential expression was run on the integrated (batch-corrected) data; and because individual cells were not linked to their CRISPRi guide, every comparison is between sorted populations. The gene lists are candidates, not tested dependencies. An exploratory three-step filter I tried in December 2024 was superseded and is not part of these results.
What the Analysis Produced
Candidate Genes for Each End Type
Gene-set comparisons between the staggered- and blunt-end conditions, giving candidate DNA-repair genes for each end type.
Heat Map and Volcano Plots
Gene Ontology enrichment of those lists, shown as a heat map, and volcano plots for the staggered- and blunt-end conditions (Fig. 3.8 of her thesis, credited to me).
Validation Still Needed
The thesis proposes candidate repair genes for each end type from this single-cell analysis and notes that they still need experimental validation.
Technical Implementation
Data
- Samples: Five sample-tagged populations of HEK293T CRISPRi cells (dCas9-KRAB with a pooled DNA-repair guide library): a control, the cells that did (FZ2+) or did not (FZ2−) accurately repair a reporter cut with en1FnCas9 to leave staggered ends, and the same two populations for a reporter cut with SpCas9 to leave blunt ends (SZ2+, SZ2−)
- Platform: BD Rhapsody whole-transcriptome single-cell RNA-seq; I started from the BD pipeline's count matrices
Computational Tools
- R 4.1.0 with Seurat 4.0.4 (institute HPC): QC, normalisation, integration, clustering, UMAP and differential expression
- DAVID (web Functional Annotation Chart): Gene Ontology enrichment
- Python (pandas, NumPy, matplotlib, seaborn, matplotlib-venn, adjustText): Gene-set comparisons and figures
Methods
- Normalisation: LogNormalize
- Integration: Seurat CCA anchors (FindIntegrationAnchors and IntegrateData)
- Differential expression: Wilcoxon rank-sum test, each sorted population against the control
- Enrichment: Over-representation against DAVID's whole-genome background
Why It Matters
Cas9 enzymes leave breaks with different end structures, and how cells repair staggered versus blunt breaks had not been systematically investigated. Knowing which repair factors each end type relies on could help steer repair towards precise, template-directed outcomes in therapeutic genome editing. The candidate genes from this analysis still need experimental validation.
Collaborative Research
This was a lab colleague's PhD study; I contributed the downstream single-cell analysis and, for another chapter of her thesis, a ClinVar analysis.
Principal Investigator: Dr. Debojyoti Chakraborty, CSIR-IGIB, New Delhi
Study and Experiments: A lab colleague (PhD project; Chapter 3 of her thesis)
Earlier Analysis: An earlier single-cell analysis of the data is also acknowledged in the thesis
My Contribution: Downstream re-analysis, and the heat map and volcano plots credited in her thesis (Fig. 3.8)
ClinVar Analysis: Deletion analysis and its plots for another chapter of her thesis (Fig. 5.2)
Duration of My Part: Single-cell analysis October 2024 – January 2025; ClinVar analysis May–June 2024
Status: A manuscript from Dr. Chakraborty's group using these data is in preparation
Future Research Directions
Publication Status
A manuscript from Dr. Chakraborty's group using these data is in preparation.
Interested in DNA Repair Research?
If you're interested in collaborating on DNA repair mechanisms, single-cell genomics, or computational biology research, I'd be happy to discuss potential opportunities.