An open-source image analysis pipeline for single-cell mRNA puncta quantification and colocalization

This repository provides a reproducible, batch-analysis pipeline for quantifying mRNA puncta in confocal microscopy images, designed for experiments with excitatory (VGLUT) and inhibitory (GAD) neuron cell-type masks. The pipeline enables accurate detection, segmentation, and per-cell analysis of mRNA puncta, facilitating high-throughput analysis of RNAscope datasets.

WEKA Trainable Classifier Integration (FIJI)

The FIJI Trainable Weka Segmentation (TWS) plugin integrates the image-processing framework of FIJI with the machine learning algorithms of WEKA to enable supervised and unsupervised image segmentation. Using a limited set of user-provided pixel annotations, TWS extracts multiscale image features and trains a classifier that can be interactively refined to segment complex biological images.

The classifier model in this repository ( Google Drive download link ) was trained on ~15 INHBA mRNA puncta images with 3 training (or classification) labels: puncta, background, and noise.

Post-processing of WEKA classification results (batch)

Apply watershed to segment clumps of detected puncta (in FIJI: Process -> Binary -> Watershed, or use a .macro automation for a folder of images). Before applying the watershed, ensure the output is binarized (in FIJI: Image -> Threshold -> B&W).

Custom MATLAB GUI for quantification & colocalization metrics

• Creates count masks for puncta quantification and exports numeric data (CSV), including puncta area and total count per image.

• Loads folders of excitatory (VGLUT) and inhibitory (GAD) masks to compute colocalization metrics per cell, including mean puncta per cell, # non-coloc puncta, density, etc.

• Designed for batch analysis! Process large datasets of images reproducibly and efficiently using the "batch count" / "batch coloc" buttons. Numeric info is exported to CSV for each image in the dataset.

EXAMPLE WORKFLOW

1. Adjust original mRNA images in FIJI The purpose of this step is to auto-adjust brightness/contrast of vGLUT / Gad mRNA original readouts from RNAscope so that they can be run through the Weka trainable classifier model.

• Open FIJI
• Process -> Batch -> Macro
• Run the following macro on input folder of original mRNA image data to auto-adjust brightness:

2. Classify puncta with trainable classifier (WEKA) Download example model (trained on RNAscope images for INHBA) here: Google Drive download link, or train your own in FIJI -> Plugins -> Trainable WEKA Segmentation

• Once WEKA app is loaded, click "Load Classifier" -> download and choose the pretrained model above
• "Apply Classifier" -> select only 8-10 raw images at a time (based on computer memory! FIJI will crash!)
• Popup will ask if you want results stored locally instead of opened in Fiji. When prompted, click yes and create a results folder for that batch for FIJI to store the classification results in.
• Popup will ask if you want to generate a “Probability Map,” click yes
• Let it run (approx 2 mins)! When it is finished (see log), ensure probability maps saved to results folder
• Repeat in batches until all images have been classified RESULT: Probability maps of detected puncta in the original mRNA images!

3. Threshold & segment classification results The purpose of this step is to turn the 3-layer probability map (one layer per training class: noise, background, and puncta) into a binary B&W image that we can watershed to segment clumped puncta. Essentially, “flattening” the model’s output for further analysis in MATLAB.

• Open FIJI
• Process -> Batch -> Macro
• To generate the macro, I applied the following to the classified images generated by Weka: Image -> Adjust -> Threshold (dark background, B&W) -> Apply -> Convert to Mask Process -> Binary -> Watershed Save image
• Emulate these steps to write / record the macro and batch process classifier result images, then click “Process” NOTE: save images to a new output folder path, choose input folder as the classification output images

4. Quantify / colocalize puncta in MATLAB GUI

• Download latest GUI from this repo and run in MATLAB
• Click “Load Puncta,” open folder of generated “ready-to-count” puncta masks
• Click “Load Cells,” open folder of vGLUT / Gad cell masks
• Set min px = 30 (filter outlier specks)
• Click “Count All Puncta” for one image OR “Batch Count” for all at once
• Click “Batch Coloc” to colocalize all at once
• Export CSV and graph results!