Using Support Vector Regression in multi-target prediction of drug toxicity

Abstract

We consider the task of drug activity prediction, specifically we predict the toxicity of fullerene-based nanoparticles in interaction with 1117 proteins. We use a multi-target Support Vector Regression model with a greedy feature selection technique to achieve RMSE of 362.9 on a test set. We also demonstrate the impact of hyperparameter tuning on model performance.