[SENescent Subtype Classifier based on Observable Unique phenoTypes]

This repository contains the data and code associated with the study titled “Single-cell morphology encodes functional subtypes of senescence in aging human dermal fibroblasts,” available at https://www.biorxiv.org/content/10.1101/2023.05.17.541204v1.

Paper: https://www.biorxiv.org/content/10.1101/2023.05.17.541204v1

Use Guide:

Manuscript and Figures Folder
Contains paper and associated figures
Codes Folder
Contains example workflow
Preprocessing delineates log normalization and standard scaling of morphological data
Senescence UMAP_KMEANS describes 2D dimensionality reduction and identification of KMEANS clusters
Biomarker Imputation Platform describes imputation platform using morphology to predict un-stained biomarker expression
Recombat describes reComBat batch correction for biomarkers between biorepeats
Time Series KMEANS describes time series kmeans for cell morphological trajectories
Senotherapy Response Dendrogram describes hierachal clustering to identify the 4 response modalities for single cell responses.
Xception Data Creator workflow to get single cell image instances for Xception senescence model
Xception Training Model and Applier workflow to train Xception senescence model and applying to new dataset
Cell Profiler
Contains example Cell Profiler pipeline for segmentation and IF quantification
Models
Contains trained computational models used for the SenScout framework
Segmented Images
Example segmented images of dermal fibroblasts from 45 year old patient at baseline and after DOX senescence induction
Data Availability
Please contact authors for data availability

Contents

# Cell Profiler: An example segmentation pipeline for measuring morphological features from microscopy images, with an image included in the pipeline.

# Codes: Includes Jupyter notebooks demonstrating the analysis workflows.

# Manuscript: Contains the research paper and associated figures.

# Data: Associated cvs files to see workflow results through the complete analysis.

# Models: Pre-trained models for senescence classification.

Descriptions

* #Cell Profiler: A widely available segmentation and analysis software, used here on microscopy images to achieve masks of individual cells.

* * Cell Profiler Example Segmentation Pipeline.cpproj is a CellProfiler file that can be used to extract morphological features of shapes and generate masks from a raw microscopy image. It classifies the images per channel, identifies nuclei, propogates outward to find the cytoplasm, and reads pixel intensity values within those shapes for the biomarker channels. It then produces a binary mask of nuceli and cytoplasms, and finally a csv of all the features of all the cells in the images. The masks where then fed into an external pipeline called HTCP cited in the paper, and the naming convention of features is atached in the Filtered Morphological Parameters csv file.

* #Codes: This folder contains Jupyter notebooks that outline the analysis workflows used in the study. MAtching the order of appearance in the manuscript.

* * Preprocessing.ipynb: Details the steps for log normalization and standard scaling of morphological feature data.

* * Senescence UMAP_KMEANS.ipynb: Describes 2D dimensionality reduction of the features measured using UMAP and identification of clusters using K-means clustering.

* * Xception Data Creator.ipynb: Workflow to extract single-cell image instances for the Xception senescence model.

* * Xception Training Model and Applier.ipynb: Workflow to train the Xception senescence model and apply it to new datasets. This model provides a score for a cell from 0 to 1 for how likely it is to be predicted as senescent.

* * Biomarker Imputation Platform.ipynb: Implements an imputation platform using morphology to predict unstained biomarker expression. It infers the intensity values of a biomarker channel that was not imaged based on similarities of the cells morphology to cells where we do know the biomarker intensities.

* * Recombat.ipynb: Applies reComBat batch correction for biomarkers between biological repeats.

* * Time Series KMEANS.ipynb: Analyzes time series data using K-means clustering to study cell morphological trajectories over time.

* * Senotherapy response dendrogram clustering.ipynb: Performs hierarchical clustering to identify the four response modalities for single-cell responses after drug addition.

* #Manuscript: The text and figures as it appears in publication for reference.

* * Manuscript and Figures Manuscript.pdf: The full research paper detailing the study.

* #Data: The data folder includes a subsampled quantification of cells.

* * Sampled_Dataset_Batch_Corrected_UMAPKMEANS_ZSCORE_logBM_transformed.csv: Contains all measurements of a subsampled group of cells over all conditions (Non senescent and all inducer), and includes biomarker readouts (e.g., p21, p16, β-Galactosidase, HMGB1, LMNB1) for each cell. Columns generally include - labels for grouping, Cell Profiler morphological features, HTCP morphological features, biomarker expression, batch corrected biomarker expression, UMAP coordinates, and KMEANS classification.

* #Models: This folder contains pre-trained models used for senescence classification.

* * weights.h5: Pre-trained Xception model weights for classifying senescent cells based on single-cell images.

* * StandardScaler.sav: A standard scaler model applied to all cell morphological and biomarker data values.

* * model_250.json: The complete trained model for senescence classification.

* * KMEANS.sav: Pre-fit KMEANS embeddings for clustering on morphological data.

Data Availability All data supporting the findings of this study are available within this repository. For access to large image files and movies, please contact the authors directly.

Key Information Sources Cell Morphology and Biomarker Data: Derived from primary human dermal fibroblasts obtained from donors aged 23 to 89 years. Cells were subjected to various senescence induction methods, including Bleomycin (BLEO), Doxorubicin (DOX), Atazanavir (ATV), Hydrogen Peroxide (H₂O₂), and replicative senescence (serial passaging). Senescence Subtypes: Identified through K means clustering analysis of morphological features, resulting in eleven distinct clusters (C1–C11), with C7, C10, and C11 characterized as bona fide senescence subtypes as per the >0.9 score from the Xception network. Senotherapeutic Treatments: Single-cell responses were profiled following treatments with Dasatinib + Quercetin (D+Q), Metformin, Navitoclax, Fisetin, and ARV-825.

Code/Software The analysis was performed using Python 3.8 with the following libraries: tensorflow-gpu==2.5.0 scipy==1.7.1 sklearn==1.1.1 pandas==1.4.3 matplotlib==3.4.2 seaborn==0.12.0 plotly==5.10.0 trackpy==0.5.0 umap-learn==0.5.3 numpy==1.22.4

Detailed instructions for setting up the environment and running the analysis workflows are provided in the respective Jupyter notebooks within the Notebooks folder.