[Generated by Task Expert Agent]

Cell Culture Feature Extraction

Performs multi-model sigmoid fitting on growth curve data to extract biological features.

Description

This task fits multiple growth models to time-series data (e.g., biomass vs temperature/time) and extracts key biological parameters and growth intervals.

Supported Models (6)

1. Logistic (4 parameters): Classic sigmoid growth
2. Gompertz (4 parameters): Asymmetric growth with lag
3. Modified Gompertz (4 parameters): Alternative Gompertz formulation
4. Richards (5 parameters): Generalized logistic with shape parameter
5. Weibull Sigmoid (4 parameters): Weibull-based growth curve
6. Baranyi-Roberts (4 parameters): Microbial growth model

Extracted Features

Model Parameters (with 95% CI):

• y0: Initial value
• A: Asymptotic maximum value
• mu: Growth rate parameter
• lag: Lag phase duration
• nu: Shape parameter (Richards only)

Statistical Metrics:

• R², Adjusted R², MSE, RMSE, MAE, AIC, BIC, SSE

Growth Intervals (time points at % of amplitude):

• t5, t10, t20, t50, t80, t90, t95
• Delta_t_10_90, Delta_t_20_80, Delta_t_5_95 (growth phase durations)

Dynamic Features:

• slope_max: Maximum growth rate (dy/dt)
• t_at_slope_max: Time at maximum slope
• mu_eff_max: Effective specific growth rate
• doubling_time_mid: Doubling time at mid-growth

Inputs

• data_table: Table with index column (time/temp) and one column per batch/sample

Configuration

• models_to_fit: List of models to test (default: all 6 models)

Outputs

• results_table: Table with all parameters, metrics, and growth intervals
• plots: ResourceSet containing Plotly graphs (individual + comparative plots)

Algorithm

1. Multi-start optimization: 10 initial guesses per model for robustness
2. Robust fitting: soft_l1 loss function to handle outliers
3. Confidence intervals: 95% CI using Jacobian approximation
4. Growth analysis: Numerical differentiation for slope and intervals

Notes

• All parameters are constrained to be strictly positive
• Missing values (NaN) are automatically filtered
• Outer join merges different time points across samples
• Best fit selected based on lowest SSE across multi-start runs