2nd CMI-PB Prediction Challenge Team Information Team Advisor: Barry Grant, Jason Hsiao Team Members: Peng Cheng, Javier Garcia, Brian Qian, Weikang Guan

Overview The 2nd CMI-PB Prediction Challenge is a systems immunology competition focused on analyzing longitudinal immune response data obtained through multi-omics experiments. The goal is to establish computational models that predict vaccination outcomes based on the baseline state of the vaccines.

Project Structure This project consists of five Jupyter notebooks that handle various aspects of the data processing, model training, prediction, and visualization.

Notebooks 00_data_pipeline.py Purpose: This script connects to the CMI laboratory database to fetch the latest training dataset and prediction target data. Usage: Run the script to update the local dataset with the latest information from the database. Functions: Connect to the database Fetch the latest data Save the data locally for further processing By running the script using your local python version, you can download all of the data files onto the data directory of this project. Every time the script is ran, it will purge any files that exist in that directory.

01_data_Integration.ipynb Purpose: This notebook preprocesses and integrates the fetched data, preparing it for model training. Usage: Run all cells to clean and integrate the data. Steps: Load the data Clean and preprocess the data Integrate different data sources Save the cleaned and integrated data

02_modeling.ipynb Purpose: This notebook trains machine learning models on the integrated dataset. Usage: Run all cells to train the models and select the best one for each prediction task. Steps: Load the integrated data Perform feature selection Train multiple models Evaluate models and select the best one Save the trained models and evaluation metrics

03_Prediction.ipynb Purpose: This notebook generates predictions using the trained models on the test dataset. Usage: Run all cells to make predictions and prepare the submission files. Steps: Load the test data Preprocess the test data Load the trained models Make predictions for each task Rank the predictions Save the prediction results

04_dashboard.ipynb Purpose: This notebook reads the training and prediction results and displays them on a dashboard for visualization. Usage: Run all cells to visualize the results. Steps: Load training and prediction results Create visualizations (e.g., plots, charts) Display the results on the dashboard

How to Run the Project

1. Clone the repository

2. Navigate to the project directory

3. Install the required packages: pip install -r requirements.txt

4. Run the notebooks: Open each file in your preferred Jupyter environment (e.g., Jupyter Notebook, Google Colab). Execute the cells in the following order: 00_data_pipeline.py 01_data_Integration.ipynb 02_modeling.ipynb 03_Prediction.ipynb 04_dashboard.ipynb

Detailed Steps for Running Each Notebook 00_data_pipeline.py Connect to the CMI laboratory database. Fetch the latest training data. Save the fetched data to a local file. if you're working on a file in a different directory, you will need to change ops.path.abspath('../your/intended/path/') to the intended relative path directory_path = os.path.abspath('../') if directory_path not in sys.path: sys.path.append(directory_path) import api_requests Usage： api_requests.create_subdirectory()

01_data_Integration.ipynb Load the fetched data from the local file. Clean and preprocess the data (e.g., handle missing values, normalize data). Integrate different data sources (e.g., combine datasets from different experiments). Save the cleaned and integrated data to a local file.

02_modeling.ipynb Load the integrated data from the local file. Perform feature selection (e.g., select the most relevant features for each task). Train multiple machine learning models (e.g., Random Forest, Lasso, SVR). Evaluate the models using cross-validation and select the best model for each task. Save the trained models and evaluation metrics to local files.

03_Prediction.ipynb Load the test data from a local file or a remote source. Preprocess the test data (e.g., handle missing values, normalize data). Load the trained models from local files. Use the trained models to make predictions for each task. Rank the predictions and prepare the submission files. Save the prediction results to local files.

04_dashboard.ipynb Load the training and prediction results from local files. Create visualizations (e.g., bar plots, line charts) to display the results. Display the visualizations on a dashboard for easy analysis. Results and Evaluation The results of the prediction tasks will be evaluated by the CMI-PB team. The final submission should include the ranked predictions for each task, which will be used to determine the accuracy and performance of the models.

Acknowledgements We thank the CMI-PB team for organizing this challenge and providing the datasets. We also thank our advisors, Barry Grant and Jason Hsiao, for their guidance and support throughout the project.

Contact For any questions or issues, please contact pec016@ucsd.edu; brqian@ucsd.edu; weguan@ucsd.edu; jag043@ucsd.edu.