[TensorFlow] Protein Interface Prediction using Graph Convolutional Networks

Unofficial TensorFlow implementation of Protein Interface Prediction using Graph Convolutional Networks [1].

Usage

$ unzip dataset.zip
$ python run.py

Dataset

Number of samples

Feature description

Graph
│
├── Vertex: 70 features
│    ├── Windowed Position Specific Scoring Matrix: 20
│    ├── Relative Accessible Surface Area: 1
│    ├── Residue Depth: 2
│    ├── Protrusion Index: 6
│    ├── Hydrophobicity: 1
│    └── Half Sphere Amino Acid Composition: 40
└── Edge: 2 features
     ├── Average Atomic Distance: 1
     └── CCαO Angle: 1

Result

• Refer that the results shown below are temporary results. (AUROC: 0.81421)
• The model tuning process is not finished completely.

Requirements

• TensorFlow 2.3.0
• White Box Layer 0.1.7

Contribution

• YeongHyeon Park : PIPGCN implementation (including neural network, data management)
• Myung Jin Kim and Yeong Beom Kim : review (double-check)

Reference

[1] Fout, Alex M. Protein interface prediction using graph convolutional networks. Diss. Colorado State University, 2017.
[2] Supplementary Data for NIPS Publication: Protein Interface Prediction using Graph Convolutional Networks. https://zenodo.org/record/1127774#.WkLewGGnGcY