RetroSynthNet: A Sequence-to-Sequence Model for Retrosynthesis

The optimization of drug synthesis is a persistent challenge in pharmaceutical research, driven by the need to accelerate discovery while reducing costs and experimental effort while maintaining safety and efficiency of the drug. This prompt us to come up with the idea of building a framework model RetroSynthNet, which is a streamlined sequence-tosequence model designed for retrosynthetic route prediction using the USPTO dataset. Our approach leverages an attention mechanism and a custom SMILES tokenizer to convert input molecular structures into viable synthetic routes efficiently. It was evaluated on metrics such as BLEU scores, Tanimoto similarity, and chemical validity, it demonstrated a good performance while maintaining computational efficiency. This research provides a practical tool for fast-tracking drug development and lays the foundation for future improvements in multi-step retrosynthesis and enhanced chemical prediction.