gws_plate_reader

Overview

The Cell Culture Application is a comprehensive analysis platform designed for fermentation and cell culture data analysis. It provides an intuitive workflow for processing, visualizing, and analyzing biological data from fermentation experiments.

Main Workflow Steps

1.Overview 📋

The starting point of your analysis. This step displays the results of the data loading scenario and helps you verify data quality.

What you'll see:

1. Basic Statistics (4 key metrics):Total Samples: Total number of batch-sample pairs found in info fileValid Samples: Number of complete samples (all data types present)Completion Rate: Percentage of samples with complete dataData Tables: Number of individual tables in the loaded ResourceSet
2. Missing Data Visualization (Venn Diagram):Interactive 3-circle Venn diagram showing data coverageBlue circle (Info): Samples with info file dataGreen circle (Raw Data): Samples with time series dataPurple circle (Follow-up): Samples with follow-up measurementsCenter overlap: Complete samples with all three data typesNumbers show count of samples in each regionExpandable table below lists all missing data details (Batch, Sample, Missing Value types)
3. Complete Data Visualizations:Batch Distribution: Pie chart showing sample count per batchMedium Distribution: Bar chart showing sample count per medium type

Key Actions:

• Verify all expected samples were loaded
• Identify which samples have missing data (info/raw_data/follow_up)
• Check batch and medium distribution
• Confirm data completeness before proceeding to selection

2.Selection ✅

Interactive tool to select which batch-sample pairs you want to analyze.

What you'll see:

1. Existing Selections (if any):List of previously created selections with their ID and statusExpandable view to see selection details
2. Create New Selection:Interactive Data Table: Shows all valid samples (only complete data, no missing values)Columns displayed: Batch, Sample, MediumMulti-row selection mode: Click on rows to select/deselectSelected rows are highlighted

How to use:

1. Review the table of valid samples
2. Click on rows to select the batch-sample pairs you want to analyze
3. Click "Validate Selection" button to create a new selection scenario
4. The system launches a filtering scenario that creates a new ResourceSet with only selected samples

When to Use:

• Remove failed or problematic experiments from analysis
• Focus on specific batches or conditions
• Create multiple selections for comparison (e.g., "control group", "treatment group")
• Reduce dataset size for faster analysis

Output:

• New selection scenario created and launched
• Filtered ResourceSet containing only selected batch-sample pairs
• Selection saved with timestamp for future reference

3.Data Visualization

3.1. Table View 📊

Browse your data in tabular format.

Features:

• Sortable and filterable columns
• Search functionality
• Export to CSV
• View metadata (batch, sample, medium composition)
• Pagination for large datasets

3.2. Graph View 📈

Visualize time series data with interactive plots.

Features:

• Multi-sample Plots: Compare multiple fermentation curves
• Parameter Selection: Choose which measurements to display (OD, pH, glucose, etc.)
• Color Coding: Automatically color by batch, sample, or medium type
• Interactive Zoom: Focus on specific time ranges
• Export Options: Download plots as PNG or SVG

Common Visualizations:

• Growth curves (OD vs Time)
• Substrate consumption
• Product formation
• pH evolution
• Multi-parameter overlay

3.3. Medium View 🧪

Explore and manage medium composition data.

Features:

• Medium Composition Table: View all medium formulations
• Compare Formulations: Side-by-side comparison of different media
• Import/Export: Upload new medium compositions or export existing ones
• Metadata Integration: Link medium data with fermentation results

Typical Use Cases:

• Document medium variations
• Identify formulation differences
• Prepare data for predictive analysis

4.Quality Check 🔍

After visualizing your data, validate and clean it.

Features:

• Outlier Detection: Identify and remove statistical outliers using configurable thresholds
• Data Validation: Check for missing values, duplicates, and inconsistencies
• Visual Quality Reports:Distribution plotsTime series plots with outliers highlightedMissing data visualization
• Distribution plots
• Time series plots with outliers highlighted
• Missing data visualization
• Quality Metrics:Percentage of valid data pointsOutlier statistics per sampleData completeness scores
• Percentage of valid data points
• Outlier statistics per sample
• Data completeness scores

Configuration Options:

• Z-score threshold for outlier detection
• Minimum data points per sample
• Missing value handling strategies

Output: Cleaned and validated dataset ready for analysis

5.Exploratory Analysis

5.1. Medium PCA 🔬

Principal Component Analysis on medium composition data.

Purpose: Reduce dimensionality of medium composition data and identify key formulation patterns.

Features:

• Component Selection: Choose number of principal components (typically 2-3)
• Variance Explained: Understand how much variation each component captures
• 2D/3D Visualization: Interactive scatter plots colored by batches or outcomes
• Loadings Plot: See which medium components contribute most to each PC

Insights:

• Identify similar medium formulations
• Detect clustering patterns
• Understand which nutrients drive variability

5.2. Medium UMAP 🗺️

UMAP (Uniform Manifold Approximation and Projection) for medium composition.

Purpose: Non-linear dimensionality reduction for complex medium composition patterns.

Configuration:

• Number of Neighbors: Controls local vs global structure (default: 15)
• Minimum Distance: Affects point clustering tightness
• 2D/3D Output: Choose dimensionality for visualization
• K-Means Clustering: Optional automatic clustering

Advantages over PCA:

• Better preserves local structure
• More effective for non-linear relationships
• Clearer visual separation of groups

Results:

• Interactive 2D/3D plots
• Cluster assignments (if enabled)
• Downloadable coordinates table

6.Feature Extraction 📉

Extract biological characteristics from growth curves.

Purpose: Convert raw time-series data into biological features for predictive modeling.

Extracted Features:

• Growth Parameters:Maximum growth rate (μmax)Lag phase durationExponential phase durationMaximum OD reached
• Maximum growth rate (μmax)
• Lag phase duration
• Exponential phase duration
• Maximum OD reached
• Substrate Consumption:Consumption rateYield coefficients
• Consumption rate
• Yield coefficients
• Product Formation:Production rateFinal titerProductivity
• Production rate
• Final titer
• Productivity

Configuration:

• Select which parameters to extract
• Define calculation windows
• Choose smoothing methods

Output:

• Table with one row per sample
• Columns for each extracted biological feature
• Ready for machine learning analysis

7.Advanced Analysis

These analyses require the extracted features from Step 6 (Feature Extraction) to work properly. Make sure to run Feature Extraction before using these tools.

7.1. Metadata-Feature UMAP 🎯

Combine medium composition and extracted features for comprehensive UMAP analysis.

Prerequisites: Requires extracted features from Feature Extraction step.

Purpose: Visualize relationships between medium formulations and biological outcomes.

Features:

• Combined Dataset: Merges medium metadata with extracted features
• Column Selection: Choose which features to include in UMAP
• Medium Name Coloring: Color points by medium composition
• Hover Data: Display batch, trial, and other metadata on hover
• 2D/3D Visualization: Interactive plots with clustering option

Use Cases:

• Identify medium-performance relationships
• Find optimal formulation clusters
• Detect outlier experiments

7.2. PLS Regression 📊

Partial Least Squares regression for predictive modeling.

Prerequisites: Requires extracted features from Feature Extraction step.

Purpose: Model relationships between medium composition (X) and biological characteristics (Y).

Workflow:

1. Select Target Variable: Choose which biological feature to predict (e.g., μmax, final titer)
2. Configure Model:Number of components (with cross-validation)Columns to exclude
3. Train Model: Automatic train/test split
4. Evaluate Results

Results Display:

• Performance Metrics:R² scoreRMSE (Root Mean Square Error)MAE (Mean Absolute Error)
• R² score
• RMSE (Root Mean Square Error)
• MAE (Mean Absolute Error)
• Predictions vs Actual: Scatter plots for train and test sets
• VIP Scores: Variable Importance in Projection (VIP > 1 = important)
• Component Selection: Cross-validation plot to choose optimal components

Advantages:

• Handles correlated predictors well
• Works with small sample sizes
• Provides interpretable variable importance

7.3. Random Forest Regression 🌲

Ensemble learning for non-linear predictive modeling.

Prerequisites: Requires extracted features from Feature Extraction step.

Purpose: Predict biological characteristics using decision tree ensembles.

Configuration:

• Number of Trees: More trees = better stability (default: 100)
• Maximum Depth: Controls tree complexity (None = no limit)
• Random Seed: For reproducibility
• Target Variable: Which biological feature to predict
• Feature Selection: Choose which medium components to include

Results Display:

• Performance Metrics: R², RMSE, MAE
• Feature Importance: Bar chart showing which medium components matter most
• Predictions vs Actual: Train and test set visualizations
• Top 10/20 Important Variables: Tables with importance scores

When to Use:

• Non-linear relationships expected
• Large feature sets
• Need robust predictions
• Want feature importance rankings

7.4. Causal Effect Analysis 🔗

Identify cause-and-effect relationships between medium components and outcomes.

Purpose: Go beyond correlation to understand causal relationships.

Methods:

• Causal inference algorithms
• Intervention analysis
• Counterfactual reasoning

Features:

• Causal graph visualization
• Effect size estimation
• Confidence intervals

Output:

• Directed causal graphs
• Effect magnitude tables
• Recommendations for medium optimization

7.5. Optimization ⚙️

Use genetic algorithms to find optimal medium composition.

Purpose: Automatically discover the best medium formulation to maximize biological performance.

Configuration:

1. Constraints Resource: Upload JSON file with component bounds{ "glucose": {"lower_bound": 0, "upper_bound": 20}, "nitrogen": {"lower_bound": 0.5, "upper_bound": 5} }
2. Optimization Parameters:Population size (default: 50)Number of iterations (default: 100)
3. Targets and Objectives:Define which features to optimize (e.g., μmax, final titer)Set minimum acceptable values for each targetAdd multiple targets (multi-objective optimization)

Algorithm:

• Genetic algorithm with mutation and crossover
• Pareto optimization for multi-objective cases
• Constraint handling

Results:

• Optimal medium formulation(s)
• Predicted performance values
• Convergence plots
• Sensitivity analysis

Use Cases:

• Design of Experiments (DoE) follow-up
• Medium formulation development
• Cost reduction while maintaining performance
• Multi-parameter optimization

Data Flow Summary

1. Load Data (from files or existing resources)
   ↓
2. Overview (verify data loaded correctly, check missing data)
   ↓
3. Selection (select batch-sample pairs to analyze)
   ↓
4. Visualization (Table/Graph/Medium views)
   ↓
5. Quality Check (remove outliers, validate data)
   ↓
6. Exploratory Analysis (PCA/UMAP on medium data)
   ↓
7. Feature Extraction (extract biological characteristics)
   ↓
8. Advanced Analysis (predictive models, optimization)

Tips and Best Practices

Data Preparation

• Ensure consistent column naming across files
• Check for missing values before analysis
• Document medium compositions clearly
• Use meaningful batch and sample names

Selection Strategy

• Create multiple selections for comparison (e.g., "good batches", "failed batches")
• Keep a "full dataset" selection for reference
• Select only valid samples (complete data)

Visualization & Quality Control

• Visualize data first to identify potential issues
• Always run Quality Check after visualization
• Review outlier detection results manually
• Document why specific samples are excluded
• Export quality reports for documentation

Analysis Workflow

• Start with exploratory analysis (PCA/UMAP) before modeling
• Extract features before running predictive models
• Use PLS for initial modeling, Random Forest for complex relationships
• Validate models with train/test splits

Optimization

• Start with wide constraint ranges, then narrow
• Use multiple optimization runs with different random seeds
• Validate optimization results experimentally
• Consider cost constraints in multi-objective optimization

Common Issues and Solutions

Problem: "No data available"

Solution: Check that Selection step completed successfully and produced a valid ResourceSet.

Problem: "Analysis failed"

Solution:

• Verify input data has required columns
• Check for missing values in critical columns
• Ensure numeric columns are properly formatted

Problem: "Model performance is poor"

Solution:

• Increase number of samples (minimum 20-30 recommended)
• Check for outliers in data
• Try different feature sets
• Consider data normalization

Problem: "Optimization doesn't converge"

Solution:

• Increase population size or iterations
• Check constraint bounds are reasonable
• Verify target objectives are achievable
• Review quality of training data

Export and Reporting

All analysis steps provide export options:

• CSV Downloads: Tables, coordinates, metrics
• Plot Exports: PNG, SVG formats
• Scenario Reports: Complete analysis documentation
• Model Artifacts: Save trained models for reuse