Title:	Drug Intraction and Side effect prediction
DNr:	Berzelius-2026-56
Project Type:	LiU Berzelius
Principal Investigator:	Golnaz Taheri <golnazt@kth.se>
Affiliation:	Kungliga Tekniska högskolan
Duration:	2026-02-27 – 2026-09-01
Classification:	10201
Keywords:

Abstract

Adverse drug reactions (ADRs) are a major challenge in drug development and clinical practice, contributing to patient morbidity, hospitalizations, and high attrition rates in pharmaceutical pipelines. Current computational approaches for side-effect prediction are limited by incomplete modeling of the complex, multi-relational interactions among drugs, proteins, pathways, and phenotypes. The goal of this project is to develop advanced machine learning models for predicting drug side effects using graph-based representation learning and large-scale multimodal data integration. We will construct heterogeneous biomedical knowledge graphs integrating drug–target interactions, protein–protein interaction networks, gene expression signatures, molecular structures, and known adverse event databases. On these graphs, we will apply and extend state-of-the-art graph neural networks (GNNs), graph transformers, and self-supervised pretraining strategies to learn biologically meaningful embeddings for drug safety prediction. Our approach leverages recent advances in deep learning, including attention-based graph models, contrastive learning, and foundation-model-inspired architectures for biomedical networks. The models will be trained on large-scale datasets (e.g., drug–target databases, adverse event reporting systems, molecular graph datasets), requiring substantial GPU resources for training, hyperparameter optimization, and ablation studies.