Dynamic Nucleic Acids Lab | Franco Lab · UCLA
Condensate Partition Coefficient Pipeline
View pipeline on GitHub →This pipeline done in python automates the measurements of condensates concentrated within nuclei compared to the surrounding dilute phase using 3D confocal Z-stacks. By processing raw multi-channel fluorescence images through automated segmentation masks, it isolates condensates and nuclei to accurately calculate partition coefficients (PC). This interactive demo showcases a precomputed sample run - reproducing manual ground-truth measurements to within ~6% - and allows you to scroll through individual slices to see and compare the raw channels and segmentation masks.
Interactive Z-Stack Viewer
Scroll through all 55 confocal slices to see, at each depth, the raw channels, the segmentation masks overlaid, and the isolated condensate and nuclei masks.
Background: condensates & RNA nanostars
Biomolecular condensates are membraneless compartments formed by liquid-liquid phase separation that concentrate proteins and RNA to regulate cellular processes. The Franco lab at UCLA engineers programmable RNA nanostars that self-assemble into synthetic condensates inside living mammalian cells, in both the nucleus and cytoplasm. The construct measured here, JABr, is a three-arm RNA nanostar with 15-nucleotide arms, kissing-loop variant A, and the Broccoli fluorescent aptamer. Broccoli lights up when it binds the dye DFHBI, so condensates are directly visible in confocal microscopy without antibodies. Imaging is of HEK293T cells across two channels: nuclei (Hoechst) and condensate (DFHBI–Broccoli).
JABr — 3-arm RNA nanostar
Three RNA strands self-assemble into a Y-shaped nanostar; sticky kissing-loop ends drive phase separation into condensates. The Broccoli aptamer makes them fluorescent under DFHBI.
The measurement: partition coefficient
The partition coefficient (PC) is the condensed-phase density divided by the dilute-phase density, after background subtraction (following the Fabrini et al. definition). It quantifies how strongly a molecule concentrates into condensates relative to the surrounding dilute phase. This automated measurement replaces a slow, subjective, manual Imaris tracing workflow that doesn't scale to large datasets.
Condensed-phase density
brightest 75% of voxels in condensate ∩ nucleus
Dilute-phase density
50 quietest 10³ patches in nucleus
Partition coefficient
PC = condensed ÷ dilute (background-subtracted)
Pipeline workflow
Raw 3D confocal Z-stacks are loaded with tifffile, then run through the steps below; partition-coefficient intensities are read from the raw (not denoised) stack.
Denoise
Cellpose 3 denoise_cyto3
Segment nuclei
Cellpose 3 cyto3 · 3D
Clean + fill voids
relabel · fill donut holes
Detect condensates
blob_log (LoG)
Intra-nuclear gate
≥50% volume in nucleus
Measure + calibrate
PC → Imaris scale
Headline Result — Sample JABr_2_5_3
The automated pipeline reproduces the manual/Imaris reference partition coefficient to within ~6%.
Raw PC
4.867
Calibrated PC
4.813
Manual reference
≈4.558
Imaris ground truth
Calibrated PC of 4.813 vs. the manual reference of ~4.558 — a deviation of about +5.6%, confirming the automated 3D pipeline agrees with hand-labeled measurements.
Key Metrics
| Metric | Value |
|---|---|
| Construct | JABr |
| Partition coefficient (raw) | 4.867 |
| Partition coefficient (calibrated) | 4.813 |
| Condensate density | 319.46 |
| Dilute density | 65.64 |
| Background | 79.0 |
| Condensates detected | 10 |
| Nuclei detected | 6 |
Precomputed sample run · pipeline not executed live
Validation & scope
Pearson r
0.942
Mean abs. error
12.9%
Within ±20%
79%
n = 28 JABr cells
The pipeline is validated and production-ready for JABr, but it is construct-specific by design: a leave-one-construct-out test showed that a single detector doesn't generalize zero-shot to unseen constructs, and fails by under-detection. Extending it to new constructs will require per-construct calibration and few-shot retraining rather than assuming a universal detector.