Cellquant: a vibecoder's guide to image analysis

Key results

1. ShownIn 3D analysis of yeast grown at 25 to 40°C, cell volume and Nsr1 intensity rose with temperature while nucleolar volume and Tif6 intensity did not track it, Sis1 intensity rose at the two highest temperatures, and Nsr1-Tif6 colocalization fell at 36°C and 40°C.

   8,309 cells from 30 biological replicates (6/5/5/6/8 per temperature), replicate-median rank tests vs 25°C, Bonferroni-corrected: cell volume median 37 µm^3 at 25°C vs 65 µm^3 at 40°C (padj = 0.003); Nsr1 intensity padj = 0.003 at 40°C; Sis1 intensity up at 36°C and 40°C (padj = 0.017, 0.035); Nsr1-Tif6 Pearson R down (padj = 0.009, 0.003). · Multi-parameter analysis of temperature…; Statistical analysis; Figure legends

2. SuggestedPrincipal component analysis (PCA) of 45 per-cell features placed temperature on PC2, while PC1 split cells into two modes: one shifted with heat and the other stayed put, and the shifting ('responsive') share dropped at 40°C.

   PC2 vs temperature ρ = +0.61 (PC1 ρ = +0.20); two-component Gaussian fit on PC1, Ashman's D = 2.62, antimode at −3.29; from 25 to 40°C the high mode moved +4.2 PC1 units (padj = 0.003), the low mode −1.0 (padj = 0.57); high-mode share fell from 84% to 64% at 40°C (padj = 0.003), on 7,569 of 8,309 cells after a 10 µm^3 volume filter. · Dimensionality reduction of single cell data…; Statistical analysis; Figure legends

3. ShownA single CLI script handles segmentation, puncta, colocalization, morphometry and proximity in 2D or 3D. By default it refuses common shortcuts: it blocks colocalization on projections and aborts 3D runs with missing, implausible or metadata-inconsistent voxel sizes, unless the user passes an explicit override flag.

   3D colocalization uses Pearson and Manders on all voxels with per-channel Costes thresholds; projection values need --allow-2d-colocalization and are flagged in provenance. 3D runs abort if XY voxel <0.01 or >10 µm, Z <0.05 or >50 µm, or values differ from metadata by >1%. provenance.json stores the CLI, SHA-256 of inputs, model checkpoints and dependency versions. · A single-script pipeline for multi-channel…; Three-dimensional segmentation and…; Metadata handling and provenance

4. SuggestedIn arsenite-treated HCT116 cells, the share of G3BP1 and PABPC1 signal in puncta rose substantially and puncta size modestly, but the experiment was too small to establish significance.

   500 µM sodium arsenite for 1 h; 79 cells from 3 control and 4 arsenite biological replicates; two-sided Mann-Whitney on replicate medians, where the smallest attainable p is 0.057, so effect sizes (fold change, Cliff's δ in Table S1) carry the result. · Stress granule quantification in human cells; Statistical analysis; Figure legends; Tables

5. ClaimedThe intended workflow is for users to describe the analysis to an LLM and iterate on QC overlays. The pipeline itself was assembled with Claude Code from existing open-source packages, without new analytical algorithms.

   Development example: 3× downsampling before Cellpose helped mammalian segmentation but destroyed spatial detail in the smaller yeast cells. The authors report the yeast analysis needed eleven lines of experiment-specific code. No evaluation with users lacking coding experience is reported. · Development of cellquant through vibecoding; Installation, tutorial, documentation; Discussion