Predicting the mechanism of action of drugs using Machine Learning
Mechanism of Action Prediction was a data science competition hosted on Kaggle by Laboratory for Innovation Science at Harvard. The goal of the competition is advancing drug development through improvements to MoA prediction algorithms.
- Data: Gene expression data and cell viability data, compound or control perturbations, time and dosage of drugs along with the targets.
- Evaluation Function: log loss.
- Problem Type: Multilabel Classification problem.
Take a look at the competition data, rules and evaluation criteria : MoA Prediction
- Overview of the data, Exploratory Data Analysis and Principal Component Analysis of MoA Prediction : MoA Prediction EDA
- Training a simple neural network model to predict MoA : MoA Neural Network
- Improving the NN model with K fold cross validation : MoA NN KFold CV
The code can be run in any of the following ways:
- Run on Kaggle: The notebooks can be found on kaggle and can be directly run on kaggle by clicking the copy and edit button on the top right corner.
- Run on Google Colab: The notebooks can be downloaded from github and run on Google Colab which has access to free cloud resources provided by Google.
- Run on local machine: The code can be downloaded or you can also clone the repository. The notebooks can be run using jupyter-notebook on your local machine. Make sure that you have all the required libraries installed.
- The simple NN model got a score of 0.02018 on public dataset and 0.01761 on private dataset.
- After KFold Cross validation, it could achieve a score of 0.01914 on public dataset and 0.01668 on private dataset.
This was the very frist time I participated in a live Data Science Competition and I did learn a lot of things on the way. Thanks to the Kaggle communtiy. Although I was just starting out in data science, I did manage to reach among the top 49% in the competition. I plan to participate in more competitions in the future.