Deep Learning for Drug Response Prediction with Gene Expression Data

Accurately predicting drug responses based on individual patient profiles is a critical challenge in personalized medicine, primarily due to the complex biological variability involved. This paper presents a deep learning framework for predicting changes in gene expression, providing insights into how drugs impact cells at the molecular level. Using data from the Kaggle competition, several models have been evaluated, including LSTM, GRU, Transformer, and Autoencoder architectures. Among these, the 3 -stacked GRU with Attention demonstrated superior performance, achieving the highest sign accuracy of 79% and the lowest mean absolute error across diverse biological conditions. The robust performance of the model highlights the effectiveness of attention mechanisms in capturing critical patterns in gene expression data.