Skip to content

README.md

Latest commit

 

History

History
97 lines (67 loc) · 3.53 KB

README.md

File metadata and controls

97 lines (67 loc) · 3.53 KB
Raw

Feature Extraction

Python Keras TensorFlow NumPy

Feature extraction pipeline for protein structure prediction using deep learning. Leverages pre-trained Keras CNN models to extract features from image datasets, with pre-computed precision matrix features for test proteins.

Overview

This repository implements a transfer learning feature extraction pipeline combining two approaches:

  1. Keras-based image feature extraction — A 3-step pipeline that builds a dataset, extracts CNN features using a pre-trained model (e.g., VGG16, ResNet), and trains a lightweight classifier on the extracted features.
  2. Pre-computed protein structure features — NumPy precision matrix files (.npy) representing structural features of protein chains, used as test inputs for downstream prediction tasks.

Repository Structure

Feature-Extraction/
├── keras-feature-extraction/       # Transfer learning pipeline
│   ├── build_dataset.py            # Build and organize image dataset
│   ├── extract_features.py         # Extract CNN features using pre-trained Keras model
│   ├── train.py                    # Train classifier on extracted features
│   └── pyimagesearch/              # Helper utilities (imutils-based)
│
└── test-features/                  # Pre-computed protein structure features
    ├── 1am9A.pre441.npy            # Precision matrix for protein 1AM9, Chain A
    ├── 1b33N.pre441.npy            # Precision matrix for protein 1B33, Chain N
    └── 1b9wA.pre441.npy            # Precision matrix for protein 1B9W, Chain A

Modules

keras-feature-extraction

A full transfer learning pipeline for image classification:

Step Script Description
1 build_dataset.py Organizes raw images into class-labeled folder structure
2 extract_features.py Passes images through a pre-trained CNN; saves extracted features to disk
3 train.py Trains a lightweight classifier (e.g., Logistic Regression) on the saved features

Why this approach? By extracting features once and saving them, the full network only runs once — dramatically reducing training time compared to end-to-end fine-tuning.

test-features

Pre-computed precision matrix (inverse covariance matrix) features derived from protein 3D structure coordinates:

File Protein Chain Shape
1am9A.pre441.npy 1AM9 A 441 x 441
1b33N.pre441.npy 1B33 N 441 x 441
1b9wA.pre441.npy 1B9W A 441 x 441

Each file stores a precision matrix capturing inter-residue correlation patterns from protein structural data.

Quick Start

# Clone the repo
git clone https://github.com/nachammai779/Feature-Extraction.git
cd Feature-Extraction

# Install dependencies
pip install tensorflow keras numpy imutils scikit-learn

# Step 1: Build dataset
python keras-feature-extraction/build_dataset.py

# Step 2: Extract features
python keras-feature-extraction/extract_features.py

# Step 3: Train classifier
python keras-feature-extraction/train.py

Requirements

  • Python 3.6+
  • TensorFlow / Keras
  • NumPy
  • scikit-learn
  • imutils