Matching Protein Sheet Partners by Feedforward and Recurrent Neural Network

1
Matching Protein ?-Sheet Partners by Feedforward
and Recurrent Neural Network

• Proceedings of Eighth International Conference on
  Intelligent Systems for Molecular Biology
  (ISMB2000), pp. 25-36
• P. Baldi, G. Pollastri, C. Anderson, and S.
  Brunak
• Cho, Dong-Yeon

2
Introduction

• Prediction of the Secondary Structure of Proteins
• Understanding their three dimensional
  conformations
• ?-helices are built up from one contiguous region
  of the polypeptide chain.
• ?-sheets are built up from a combination of
  several disjoint regions.
• Previous Studies
• The best existing methods for predicting protein
  secondary structure achieve prediction accuracy
  in 75-77 range.
• ?-sheet is almost invariably the weakest category
  in terms of correct percentages.
• Prediction of Amino Acid Partners in ?-sheets

3
Data Preparation

• Selecting the Data
• 826 protein chains from the PDB select list of
  June 1998
• Assigning ?-sheets Partners

A2-B2 A3-B3 B2-C2 B3-C3 C2-D2 C3-D3
4
Statistical Analysis

• First Order Statistics
• The frequency of occurrence of each amino acid

General amino acid frequencies in the data
Amino acid frequencies in ?-sheets
5

• The ratio of the frequencies in ?-sheets over data

6

• Second Order Statistics
• The conditional probabilities P(XY) of observing
  a X knowing that the partner is Y in a ?-sheet

7

• Logo representation

8

• Length Distribution
• Interval distances between paired ?-strands,
  measured in residue positions along the chain

9
Artificial Neural Network Architecture

• Feedforward Neural Network
• Large input windows
• They tend to dilute sparse information present in
  the input that is really relevant for the
  prediction.
• Two-window approach
• One can either provide the distance information
  as a third input to the system or one can train a
  different architecture for each distance type.

10

• The architecture
• Two input windows of length W
• The number D of amino acid is also given as an
  input unit to the architecture with scaled
  activity D/100.
• The goal is to output a probability reflecting
  whether the two amino acids located at the center
  of each window are partners or not.

11

• Recurrent Neural Network
• Bi-directional recurrent neural network (BRNN)
• Input layer
• Forward and backward Markov chain
• Output layer

12
Experiments and Results

• Data
• Randomly split the data 2/3 for training and 1/3
  for test
• Extremely unbalanced
• At each epoch, all the 37008 positive examples
  are presented with 37008 randomly selected
  negative examples.
• The total balanced percentage is the average of
  the two percentages obtained on the positive and
  negative examples.

13

• Results
• Feedforward neural network
• The best architecture

14

• The predicted second order statistics

15

• Five-fold cross validation
• BRNN Architecture
• Three values (7, 9, and 11) are used as the size
  of two input windows.
• Length 7 yields again the best performance.

16

• Five-fold cross validation
• Ensemble architecture
• The ensemble of 3 BRNNS
• Five-fold cross validation

17

• Summary of all the five-fold cross validation
  results
• Profile approach
• The profile approach was used as input to the
  artificial neural network.
• The overall performance is comparable, but not
  any better.
• Profiles may provide more robust first order
  statistics, but weaker intrasequence correlation.

18
Discussion

• We have developed a NN architecture that predicts
  ?-sheet amino acid partners with a balanced
  performance close to 84 correct prediction.
• It is insufficient by itself to reliably predict
  strand pairing because of the large number of
  false positive predictions.
• Some of directions for future work
• Profiles on the BRNNs
• Reduce the number of false positive predictions
• Improve the quality of the match
• Use of raw sequence information in addition to
  profiles
• ?-sheet predictor
• Various combinations of the present architectures