Protein Structure Prediction Using Novel Machine Learning Techniques for 9-mers

This repository contains the reproducible research code for the project “Protein Structure Prediction using Novel Machine Learning Techniques for 9-mers Dataset”, developed as part of AIT 736 – Applied Machine Learning (DL2) at George Mason University.

The project investigates protein fragment structure prediction using 9-mer sequences derived from CullPDB, leveraging multi-input deep learning architectures that combine amino acid sequence information with torsion-angle features (φ, ψ). The primary model is an LSTM-based neural network with dense layers and dropout regularization to capture sequential and non-linear structural patterns.

📌 Project Objectives

• Load, preprocess, and analyze a standardized 9-mer protein fragment dataset
• Engineer biologically meaningful features from sequence and torsion angles
• Design and train a deep learning model for protein structure prediction
• Evaluate predictive performance using quantitative metrics and visual analysis
• Provide reproducible experiments suitable for academic publication

---

🧬 Dataset

• Name: 9mers from CullPDB
• Source: UCI Machine Learning Repository
• Link: https://archive.ics.uci.edu/dataset/866/9mers+from+cullpdb
• DOI: 10.24432/C58024

Dataset Description

• ~158,000 protein fragments (9-mers)
• Derived from 3,733 non-redundant proteins
• No missing values

Features per sample:

1. Amino acid sequence (9 residues)
2. Secondary structure labels (9)
3. Phi (Φ) torsion angles ∈ [-π, π)
4. Psi (Ψ) torsion angles ∈ [-π, π)

⚠️ Dataset is NOT included in this repository.
Please download it directly from UCI https://archive.ics.uci.edu/dataset/866/9mers+from+cullpdb