The Gene Ontology (GO) has about 45,000 terms that define a wide range of molecular functions, biological processes, and cellular locations of proteins, making Automated Function Prediction (AFP) a difficult problem. To accurately predict protein functions from this broad vocabulary, a model must be able to manage the diversity and complexities of protein biology. Despite a number of efforts, AFP is still an unsolved problem. Protein Language Models (pLMs) are a powerful tool to understand and predict protein functions using learned amino acid sequences and their relationships. AFP is inherently a multi-label classification problem. Our model SeqToFunc uses a hybrid transformer model that encodes both the protein sequences and their interactions into contextually aware embeddings for protein function prediction. The interactions between GO graphs are handled using a Graph Convolutional Network (GCN). We finetuned ESM2 on the CAFA5 challenge dataset for GO term predictions and reported our results using F1 score.