Protein Secondary Structure Prediction Using ESMFold: A Deep Learning-Based Bioinformatics Web Application

Protein secondary structure prediction remains a cornerstone problem in computational biology, underpinning applications in drug discovery, genetic disease research, vaccine design, and protein engineering. Traditional approaches rely on evolutionary information derived from multiple sequence alignments and are computationally expensive. This paper presents a complete bioinformatics web application that harnesses ESMFold, a pretrained protein language model developed by Meta AI, to predict per-residue secondary structure labels — alpha helix (H), beta sheet (E), and coil (C) — directly from amino acid sequences in a zero-shot fashion, without requiring alignment databases or model fine-tuning. The system supports three input modalities: manual sequence entry, FASTA file upload, and live Protein Data Bank (PDB) ID retrieval. Prediction output is presented through color-coded sequence maps, per-residue pLDDT confidence heatmaps, structural composition charts, and interactive 3D molecular visualization. The application achieved Q3 accuracy between 81.4% and 87.2% on benchmark proteins spanning all major fold classes, consistent with published ESMFold performance. The tool serves both research and educational purposes, combining scientific rigor with an accessible and visually rich Streamlit interface.