Dear reader,

Welcome to our project repository on data collection, storage, and integration, where we investigate whether molecular structure data—represented using the Simplified Molecular Input Line Entry System (SMILES)—can be successfully combined with genome-wide gene expression signatures to form a unified dataset that reflects drug-induced cellular perturbations. Our primary objective is to assess whether the LINCS L1000 dataset, enriched with molecular embeddings, provides a solid foundation for future predictive modeling of drug responses.

The repository is organized as follows:

1. script folder Contains all data-collection scripts as well as the code used to upload data to the S3 bucket. (Note: You will need credentials similar to those in the .env.example file. The dataset consists of ~5 GB of compressed data.)

2. notebook folder Includes all preprocessing, exploratory data analysis (EDA), and predictive modeling notebooks. You will also find instructions to access the final cleaned dataset. (Training was performed on Google Colab using an A100 GPU.)

3. weights folder Stores the final trained model used in our experiments. (A Random Forest with 10 trees was used due to memory constraints.)

4. mol-insight folder Contains a web application that visualizes the results and insights produced in this project. (version with our preds on empty-branch) To launch locally:

   npm install  
   npm run dev

Best regards, Millan Das, Arthur de Leusse & Alexis Vannson