Geometric coherence of single-cell CRISPR perturbations reveals regulatory architecture and predicts cellular stress
| Authors | Prashant C. Raju |
| Year | 2026 |
| HF Upvotes | 3 |
| arXiv | 2604.16642 |
| Download | |
| HF Page | View on Hugging Face |
Abstract
Genome engineering has achieved remarkable sequence-level precision, yet predicting the transcriptomic state that a cell will occupy after perturbation remains an open problem. Single-cell CRISPR screens measure how far cells move from their unperturbed state, but this effect magnitude ignores a fundamental question: do the cells move together? Two perturbations with identical magnitude can produce qualitatively different outcomes if one drives cells coherently along a shared trajectory while the other scatters them across expression space. We introduce a geometric stability metric, Shesha, that quantifies the directional coherence of single-cell perturbation responses as the mean cosine similarity between individual cell shift vectors and the mean perturbation direction. Across five CRISPR datasets (2,200+ perturbations spanning CRISPRa, CRISPRi, and pooled screens), stability correlates strongly with effect magnitude (Spearman ρ=0.75-0.97), with a calibrated cross-dataset correlation of 0.97. Crucially, discordant cases where the two metrics decouple expose regulatory architecture: pleiotropic master regulators such as CEBPA and GATA1 pay a "geometric tax," producing large but incoherent shifts, while lineage-specific factors such as KLF1 produce tightly coordinated responses. After controlling for magnitude, geometric instability is independently associated with elevated chaperone activation (HSPA5/BiP; ρ_{partial}=-0.34 and -0.21 across datasets), and the high-stability/high-stress quadrant is systematically depleted. The magnitude-stability relationship persists in scGPT foundation model embeddings, confirming it is a property of biological state space rather than linear projection. Perturbation stability provides a complementary axis for hit prioritization in screens, phenotypic quality control in cell manufacturing, and evaluation of in silico perturbation predictions.
Engineering Breakdown
Plain English
This paper addresses a critical gap in predicting cellular responses to genetic perturbations: while CRISPR screens can measure the magnitude of transcriptomic changes, they don't capture whether cells respond coherently or scatter randomly across expression space. The authors introduce Shesha, a geometric stability metric that quantifies directional coherence by computing mean cosine similarity between individual cell shift vectors and the population's mean perturbation direction. This metric reveals that two perturbations with identical effect magnitude can produce qualitatively different outcomes—one driving cells along a shared trajectory while another scatters them—a distinction invisible to traditional magnitude-only analyses. The work bridges a fundamental blind spot in genomics: knowing how far cells move is less informative than knowing whether they move together.
Core Technical Contribution
The core novelty is a geometric stability metric (Shesha) that decouples effect magnitude from directional coherence in single-cell perturbation responses. Rather than treating each cell's expression change as an independent scalar (as prior work does), this approach embeds each perturbation response as a vector in expression space and measures how aligned individual cell vectors are with the population mean direction. The key insight is that coherence—the degree to which perturbed cells cluster along a single trajectory—is a distinct and predictively valuable dimension orthogonal to effect size. This shifts the paradigm from univariate magnitude-based metrics to multivariate geometry-aware analysis, enabling detection of perturbations that scatter cells versus those that drive coordinated state transitions.
How It Works
The input is single-cell RNA-seq data from CRISPR screens: expression vectors for unperturbed cells and their corresponding perturbed variants. For each perturbation, the method computes shift vectors by subtracting baseline expression from perturbed expression for each individual cell, producing a cloud of direction vectors in expression space. The mean perturbation direction is calculated as the average of all these shift vectors across the cell population. Shesha then computes the cosine similarity between each individual cell's shift vector and this population mean direction, averaging these similarities to produce a single coherence score between 0 and 1. A score near 1 indicates cells moved together coherently; a score near 0 indicates scattered, incoherent responses. The metric is applied across five CRISPR screen datasets to validate that coherence meaningfully stratifies perturbations independent of magnitude.
Production Impact
For genome engineering pipelines, this metric enables smarter perturbation prioritization: engineers can now identify which genetic perturbations produce robust, reproducible cellular state transitions versus noisy, unpredictable ones—critical for therapeutic design where you need precise, coordinated effects. In practice, this means your CRISPR screening workflow could rank candidates by (magnitude, coherence) pairs rather than magnitude alone, catching high-magnitude but incoherent perturbations that would fail in patient populations. Computationally, the metric is cheap: it requires only matrix operations on already-computed cell shift vectors, adding minimal overhead to existing scRNA-seq analysis pipelines and scaling linearly with cell count and gene dimensionality. Integration into standard single-cell analysis frameworks (Scanpy, Seurat) is straightforward, but requires validation that coherence scores remain stable across different batch correction methods and sequencing platforms.
Limitations and When Not to Use This
The paper assumes that coherence in expression space correlates with functional coherence in phenotypic outcomes, but this assumption isn't validated against actual cell behavior or organismal phenotypes—high geometric coherence could still produce heterogeneous functional outcomes. The metric depends heavily on accurate baseline definition and batch correction; if unperturbed cells are mislabeled or batches are poorly corrected, coherence scores become meaningless, making this approach sensitive to upstream data quality in ways not thoroughly characterized. The paper is limited to in vitro CRISPR screens; applicability to in vivo perturbations (where cell migration and selection complicate interpretation) remains unexplored. Finally, the five datasets mentioned in the abstract are insufficient to determine whether coherence is truly predictive of downstream success in genome engineering applications—this is primarily a geometric characterization without direct validation against functional outcomes.
Research Context
This work extends single-cell perturbation analysis beyond prior magnitude-centric approaches (which typically use MAST, DESeq2, or simple fold-change metrics) by incorporating geometric principles from representation learning. It builds on foundational CRISPR screening papers and recent advances in single-cell geometry analysis (work on manifold learning in scRNA-seq), recognizing that cellular state transitions are fundamentally geometric phenomena. The contribution opens a new evaluation dimension for CRISPR screens, positioning coherence as a complementary metric to effect size, similar to how variance and mean are both essential in statistics. This could catalyze follow-up work linking geometric coherence to cell-state stability, predicting off-target effects, or designing combinatorial perturbation strategies that maximize coordinated responses.
:::tip Subscribe Get weekly breakdowns of papers like this in AI Letters - the newsletter for engineers building production AI systems. :::