MOTIVATION: As most proteins interact with other proteins to perform their respective functions, methods to computationally predict these interactions have been developed. However, flawed evaluation schemes and data leakage in test sets have obscured the fact that sequence-based protein-protein interaction (PPI) prediction is still an open problem. Recently, methods achieving better-than-random performance on leakage-reduced PPI data have been proposed. RESULTS: Here, we show that the use of ESM-2 protein embeddings explains this performance gain irrespective of model architecture. We compared the performance of models with varying complexity, per-protein, and per-token embeddings, as well as the influence of self- or cross-attention, where all models plateaued at an accuracy of 0.65. Moreover, we show that the tested sequence-based models cannot implicitly learn a contact map as an intermediate layer. These results imply that other input types, such as structure, might be necessary for producing reliable PPI predictions. AVAILABILITY AND IMPLEMENTATION: All code for models and execution of the models is available at https://github.com/daisybio/PPI_prediction_study. Python version 3.8.18 and PyTorch version 2.1.1 were used for this study. The environment containing the versions of all other packages used can be found in the GitHub repository. The used data are available at https://doi.org/10.6084/m9.figshare.21591618.v3.