Introduction

This GitHub is designed to fit Grating-Coupled Interferometry (GCI) data from Creoptix WAVEsystem.

Set up the running environment

To run the Bayesian regression, we need some python packages.

• jax v0.4.33
• jaxlib v0.4.33
• numpyro v0.15.3
• pickle-mixin >= v1.0.2
• arviz >= 0.20.0
• uncertainties

If higher versions of JAX, JAXlib, and numpyro are installed, we need to check whether x64 jax.numpy can be used by executing the following code without any errors:

import jax
jax.config.update("jax_enable_x64", True)

Running test

1. Setting up the main directory

    DIR='/home/vla/python/gci'

2. Install the packages and download the GitHub repository

    git clone 'https://github.com/vanngocthuyla/gci.git'

3. Global Fitting for GCI Experiment

Assuming DIR is your working directory, you can perform global fitting for one set of GCI experiments, including the subtracted datasets (e.g., FC2-1, FC3-1, and FC4-1) from cycle 13, by running the following commands:

mkdir $DIR/example/output
python $DIR/scripts/run_fitting_GCI.py \
    --out_dir $DIR/example/output/13_Y \
    --global_fitting --fitting_subtract --return_y_offset \
    --init_niters 1000 --init_nburn 200 \
    --niters 5000 --nburn 2000 --nchain 4 --random_key 0 \
    --analyte_file $DIR/example/input/ZIKV_Subtraction.csv \
    --analyte_keys_included "Fc=2-1-13_Y Fc=3-1-13_Y Fc=4-1-13_Y" \
    --analyte_concentration_uM 10.0 \
    --calibration_file $DIR/example/input/ZIKV_DMSO.csv \
    --calibration_keys_included "Fc=2-11_Y Fc=3-11_Y Fc=4-11_Y" \
    --end_dissociation 10.0

Arguments for run_fitting_GCI.py:

• --out_dir: Specifies the directory where the results will be saved.
• --global_fitting: If set to True, performs global fitting for subtracted FCs. If False, fits each subtracted FC separately.
• --fitting_complex: If set to True, fits a complex model. Otherwise, fits a simple model.
• --fitting_subtract: If set to True, the provided data is subtracted (e.g., Fc=2-1-13). If False, the provided data is raw (e.g., Fc=2-13).
• --return_y_offset: If set to True, estimates the y_offset.
• --init_niters, --init_nburn, --niters, --nburn, --nchain:
  Define the number of MCMC samples and burn-in steps:
  • --init_niters: Initial MCMC samples for tuning.
  • --init_nburn: Burn-in samples during initial tuning.
  • --niters: Total number of MCMC samples collected.
  • --nburn: Burn-in samples during collection.
  • --nchain: Number of chains for Bayesian regression.
• --random_key: Sets the random key for Bayesian regression.
• --analyte_file: Specifies the input data file.
• --analyte_keys_included: Specifies the columns of analyzed data.
• --analyte_keys_included_FC1: Required if --fitting_subtract=False.
• --analyte_concentration_uM: Specifies the analyte concentration in µM.
• --calibration_file: Specifies the DMSO calibration file.
• --calibration_keys_included: Specifies the columns for DMSO data.
• --end_dissociation: Defines the end time point for dissociation.

Alternatively, you can adjust the script /scripts/submit_fitting_GCI.py to submit SLURM jobs.

SLURM job submission

SLURM Job Information

More details about SLURM jobs can be found via this link.

Submitting a Job for the Example Dataset

To submit a job for the example dataset, follow these steps:

1. Edit the main directory

Update the main directory in the file main_dir.txt.

2. Adjust the partition and Conda environment

Modify the partition and Conda environment settings in the script submit_fitting_GCI.py.

3. Submit the job

Run the following command to submit the job:

bash $DIR/example/run_me/run_me_submit_fitting_subtract.sh

4. Job Automation and Output Details

The code automatically detects datasets from different cycles in the input file ZIKV.csv and submits multiple jobs to the server. Each job is named according to its corresponding cycle number.

Output Information:

• All submission details are stored in .job and .log files located in the defined output directory.
• Results for each experiment fitting are saved in subfolders named after their respective cycle numbers. Each subfolder includes:
  • Autocorrelation plot
  • Trace plot
  • MCMC samples saved as traces.pickle
  • Summarized MCMC samples saved as Summary.csv
  • MAP estimates saved as map.pickle and map.csv
  • A plot of the fitted sensorgram (gray curve) overlaid with observed dissociation segment data (red dots), along with the mean and standard deviations of parameter estimates from Bayesian regression
  • A diagnostic plot for derivative and integral curves