This repository contains the code for the paper: Towards Early Prediction of Human iPSC Reprogramming Success.

• Code
  • Models
  • Annotation
  • Data processing and Evaluation
• Setup and Commands
• Data
• Trained Models
• Supplementary material
  • Visualizations
  • 3D Plots

Code

Models

The 8 models reported in the paper are located as follows:

• classification
  • XGB: ipsc_data_processing/eval_cls.py
  • SWC: ipsc_classification/swin
  • CNC: ipsc_classification/convnext
• static segmentation
  • SWD: ipsc_static_segmentation/swin_detection
  • CND: ipsc_static_segmentation/swin_detection
• video segmentation
  • IDOL: ipsc_video_segmentation/ipsc_vnext
  • SEQ: ipsc_video_segmentation/ipsc_vnext
  • VITA: ipsc_video_segmentation/ipsc_vita

The Swin transformer semantic segmentation model not reported in the paper is available in ipsc_static_segmentation/swin_semantic.

Annotation

• labeling tool used for stages 1 and 2 of the annotation process can be run using ipsc_labelling_tool/labelImg.py
• retrospective labelling for stage 3 can be run using ipsc_labelling_tool/propagate_by_tracking.py

Data processing and Evaluation

• scripts for converting annotations between various formats like XML, CSV and JSON are available in ipsc_data_processing/
• classifiers can be evaluated using ipsc_data_processing/eval_cls.py
• detectors can be evaluated using ipsc_data_processing/eval_det.py
  • this is also the script used for generating the detection failure visualization videos

Setup and Commands

Each of the above folders contains a subfolder named cmd containing markdown files with hierarchically organized list of commands to reproduce any of the results reported in the paper.

• for example, the commands for SWD are in swin_det.md and swin_det_setup.md in ipsc_static_segmentation/swin_detection/cmd while those for IDOL and SEQ are in vnext.md and vnext_setup.md in ipsc_video_segmentation/swin_detection/cmd
• some of the contents of these files might not be easy to understand at present but I am working to make these more user-friendly
• if the commands needed to reproduce any results in the paper or to run any of the models on a new dataset are not clear, please create an issue or contact me

Data

Images and annotations can be downloaded from here:

• ROI images and labels
• Raw 714 MP images
• List TXT files
• Static segmentation JSON files
• Video segmentation JSON files

ROI images and labels, list TXT files and static segmentation JSON files should be extracted to /data/ipsc/well3/all_frames_roi/ while video segmentation JSON files should be extracted to /data/ipsc/well3/all_frames_roi/ytvis19/.

Trained Models

Trained models can be downloaded from here

• both early-stage and late-stage trained models are included
• the zip file for any model should be extracted in its source directory while maintaining the folder structure inside the zip file
  • for example, the models for IDOL should be extracted inside ipsc_video_segmentation/ipsc_vnext
  • this would extract the .pth checkpoint files into a subfolder named log/idol-ipsc-ext_reorg_roi_g2_16_53 and log/idol-ipsc-ext_reorg_roi_g2_54_126 for the early and late-stage models respectively

Supplementary material

The supplementary PDF is available here

Visualizations

Videos visualizing the annotations along with detection results and failures are available here. Detailed description of these videos is in the supplementary PDF.

3D Plots

Frame-wise partial AUC plots are available here as interactive HTML files.