Protein-ligand binding affinity prediction is a central problem in computational drug discovery. Accurate prediction of binding strength can accelerate the identification and optimization of candidate compounds while reducing the need for costly experimental screening. Recent advances in machine learning, including deep learning and foundation models for proteins and small molecules, have shown promising performance on benchmark datasets. However, significant challenges remain regarding model generalization, dataset bias, robustness, and fair comparison across different methodologies. The objective of this project is to benchmark and develop machine learning approaches for protein-ligand binding affinity prediction. The work will involve large-scale evaluation of existing methods, including sequence-based, structure-based, and foundation-model-based approaches, on publicly available datasets such as PDBBind and related benchmark collections. In addition, new predictive models and representations will be investigated to improve predictive accuracy and generalization to unseen protein families and ligand chemotypes. The project requires substantial computational resources due to large-scale model training, hyperparameter optimization, and evaluation across multiple datasets and model architectures. Experiments will involve GPU-accelerated deep learning frameworks and processing of large protein and molecular structure datasets. The project is conducted within the WASP PhD programme at Chalmers University of Technology. Main supervisor: Simon Olsson, Chalmers University of Technology.