Life science practitioners are drowning in unlabeled protein sequences. Natural Language Processing (NLP) community has recently embraced self-supervised learning as a powerful approach to learn representations from unlabeled text, in large part due to the attention-based context-aware Transformer models. In a transfer learning fashion, expensive pretrained universal embeddings can be rapidly fine-tuned to multiple downstream prediction tasks. In this work we present a modification to the RoBERTa model by inputting a mixture of binding and non-binding protein sequences (from STRING database) during pre-training with the Masked Language Modeling (MLM) objective. Next, we compress protein sequences by 64\% with a Byte Pair Encoding (BPE) vocabulary consisting of 10K tokens, each 3-4 amino acids long. Finally, to expand the model input space to even larger proteins and multi-protein assemblies, we pre-train Longformer models that support 2,048 tokens. Our approach produces excellent fine-tuning results for protein-protein binding prediction, TCR-epitope binding prediction, cellular-localization and remote homology classification tasks. We suggest that the Transformer's attention mechanism contributes to protein binding site discovery. Further work in token-level classification for secondary structure prediction is needed.