Multi-Cell-Line-Learning

This repository contains codes and data for the MCL paper. The following are the instructions to run the codes.

Citation

Coming soon

Requirements

Dependencies can be found in the environment.yml file.

⚠️ It is recommended to run IPOPT with HSL solvers here.

Usage

Prepare dataset

1. Split datasets into the training and testing set and fill data into the template files, Data_input_train.xlsx and Data_input_test.xlsx.

   ⚠️ Notation change, $\beta$ of enzymes are equivailent to E0d columns in the excel file.

2. Provide feasbile solutions of each steady-state time point to create warmstart points for the algorithm. The input template can be found in \SCL_MCL\Data\IGs\CL1\solution_CL1_1.csv. The filename (path) should follow the following template \SCL_MCL\Data\IGs\$cell-line-name\solution_$cell-line-name_$random-number.csv.

Steps to run SCL/MCL optimization codes in the SCL_MCL folder

1. Solve SCL for each cell line

   python Wrapper.py -stg 1 -ti $cell-line-name
   

   Run the code with the specified $cell-line-name.

2. Perform 5-fold cross-validation to tune lambda in MCL

   For each $\lambda$ value, change the input values of -l1 in the following code in the log scale, i.e. -1 for $\lambda = 0.1$.

   python Wrapper.py -kidx $kidx -l1 $l1 -np 16 -stg 2
   

   Run the code with -kidx from 1 to 5. Then, run the following code to do the model validation on the validation set.

   python Wrapper.py -kidx $kidx -l1 $l1 -np 16 -stg 3
   

3. Solve MCL on the full dataset

   Based on the previous stage, pick up the optimal $\lambda$ value and run

   python Wrapper.py -l1 $l1 -np 16 -stg 4
   

4. Model Validation on the testing data set

   Get the testing error for SCL:

   python Model_Validation.py -stg 1
   

   Get the testing error for MCL:

   python Model_Validation.py -l1 $l1 -stg 2
   

Steps to run the final cell-line models based on MCL in the Final_Model folder

1. Move MCL solutions to input\Model_inputs_MCL.csv
2. Run the steady-state simulation using the specified cell line and flux-state model

   python ModelWrapper.py -CL CL1 -fs HF -mu_target 1.0 -gln_target 2.5 -glcUB 30 -glcLB 5 -lacUB 15 -lacLB 2.0 -step_number 20 -init_step_size 0.5 -stop_step_size 1e-1 -fname SS_sol
   

   Read the ModelWrapper.py file for the details of the input arguments.