README.md

ANN-for-protien-secondary-structer

---

Approximately replecated methodology from ...

Using knowledge-based neural networks to improve algorithms:
Refining the Chou-Fasman algorithm for protein folding

---

Data Set:

Test set for study of secondary structure of globular proteins by Ning Qing and Terry Sejnowski.

There were 128 protiens in the UC-Irvine archive. Train and Test were combined into one file (with train in front). Data was split into three test sets for our version of this network: Train, Tune and Test.

Out of 128 proteins, they were broken up as follows:

• tune: index % 5 = 0
• train: index % 6 = 0
• test: the rest of the proteins.

---

Network:

This network is implemented with:

• Hinton's Droput
• Momentum Term for backpropagation
• Early Stopping

Network preforms best with the configuration found in the main method.

---