Transforming Protein Secondary Structure Prediction with Deep Learning: A Novel Approach Using Encoder-Decoder Architectures

Predicting protein secondary structure is a pivotal challenge in structural biology, given its critical role in understanding protein function and designing novel therapeutics. Traditional methods, while informative, fall short of providing the accuracy and efficiency required for real-world applications. This thesis introduces an innovative application of Encoder-Decoder neural networks, originally designed for Natural Language Processing, to advance protein secondary structure prediction. By treating amino acid sequences as structured data akin to language, we leverage embeddings and sequential modeling to capture intricate relationships within the protein chain. We present three novel deep learning architectures tailored for this task and evaluate them on a custom-built dataset derived from the Protein Data Bank. Our approach demonstrates the potential to enhance Q8 accuracy and provide reliable preliminary predictions for experimental biologists. By bridging cutting-edge AI techniques with the demands of structural biology, this work aims to set a new benchmark for computational tools in protein research.