profiles and fuzzy k-nearest neighbor algorithm for protein secondary structure prediction rajkumar...

Profiles and Fuzzy K-Nearest Neighbor Algorithm for Protein Secondary Structure Prediction

Rajkumar Bondugula,

Ognen Duzlevski and Dong Xu

Digital Biology Laboratory, Dept. of Computer Science

University of Missouri – Columbia, MO 65211, USA

Outline

Introduction Protein secondary structure prediction Popular methods K-Nearest Neighbor method Fuzzy K-Nearest Neighbor method

Methods Filtering the prediction Results and discussion Summary and Future work

Introduction

Goal: Given a sequence of amino acids, predict in which one of the eight possible secondary structures states {H, G, I, B, E, C, S,T} will each residue fold in to.

CASP convention {H,G,I} → H {B,E} → E {C,S,T} → C

Example:Amino Acid VKDGYIVDXVNCTYFCGRNAYCNEECTKLXGEQWASPYYCYXLPDHVRTKGPGRCHSecondary StructureCEEEEEECCCCCCCCCCCHHHHHHHHHHCCCCEEEECCEEEEECCCCCCCCCCCCC

Protein 3-Dimensional structure

Importance of Secondary Structure

An intermediate step in 3D structure prediction structure → function

ClassificationEx: α, β, α/β, α+β

Helps in protein folding pathway determination

Existing Methods

Popular MethodsNeural Network methods

Ex: PSIPRED, PHD

Nearest Neighbor methods Ex: NNSSP

Hidden Markov Model methods

Why K-Nearest Neighbors method?

Methods based on Neural Networks and Hidden Markov models perform well if the query protein have many homologs

in the sequence databasenot easily expandable

The 1-Nearest Neighbor rule is bound above by no more than twice the optimal Baye’s error rate [Keller et. al, 1985]

K-NN will work better and better as more and more structures are being solved

K-Nearest Neighbor Algorithm

Instances to be classified Classified instances

Instances to be classified class B class F

Advantages of Nearest Neighbor methodsSimple and transparent model

New structures can be added without re-training

Linear complexity

DisadvantageSlower compared to other models as processing is

delayed until prediction is needed

Why Fuzzy K-NN?

Disadvantages of Crisp K-NN Atypical examples are given as much as weight as those that

truly represent a particular class

Once instance is assigned to a class, there is no indication of its “strength” of its membership in that class

- - - N L G A G N S G L N L G H V A L T F

- - - N L G A

- - - N L G A- - N L G A G

- - - N L G A- - N L G A G- N L G A G N

- - - N L G A- - N L G A G- - N L G A NN L G A G N S

- - - N L G A- - N L G A G- N L G A G N SL G A G N S G

Position Specific Scoring Matrix

. . . 5 -2 -2 -2 -1 -1 -1 0 -2 -2 -2 -1 -1 -3 -1 1 0 -3 . . .

. . . -1 -3 -2 -2 -3 -2 -2 -3 -3 -3 -3 -1 -3 -4 8 -1 -2 -4 . . .

. . . 3 -2 -1 -2 -1 -2 -2 2 -2 -2 -2 -1 -2 -3 -2 0 3 -3 . . .

. . . 2 -3 -3 -3 -2 -2 -3 -2 -3 1 0 -2 0 4 -3 -1 -1 -2 . . .

. . . 0 -2 0 -1 -1 -1 -1 -1 -2 -2 -2 -1 -2 -3 -2 4 4 3 . . .

. . . -1 -3 -4 -4 -1 -3 -3 -4 -4 2 0 -3 0 -1 -3 -2 -1 3 . . .

. . . 0 -2 0 -1 -1 -1 -1 -1 -2 -2 -2 -1 -2 -3 -2 4 4 3 . . .

. . . -1 -3 -3 -2 -3 -2 -2 -3 -3 -3 -4 -2 -3 -4 8 -1 -2 4 . . .

. . . 4 -2 -1 -2 -1 -1 -1 -1 -2 -2 -2 -1 -2 -3 -1 3 0 3 . . .

. . . 0 -1 0 -1 -1 -1 -1 -1 -2 -2 -3 -1 -2 -3 -1 4 4 3 . . .

. . . 0 -3 -1 -2 -3 -2 -3 6 -3 -4 -4 -2 -3 -4 -3 -1 -2 3 . . .

. . . 0 -3 -4 -4 -2 -3 -3 -3 -3 1 5 -3 1 0 -3 -2 -2 2 . . .

. . . 2 -1 0 -1 -1 -1 -1 -1 -2 -3 -3 -1 -2 -3 -1 5 1 3 . . .

. . . -2 -2 3 6 -4 -1 1 -2 -1 -4 -4 -1 -4 -4 -2 -1 -1 5 . . .

. . . 2 -3 -3 -3 -2 -2 -3 -2 -3 1 0 -2 0 4 -3 -1 -1 2 . . .

. . . 0 -2 0 -1 -1 -1 -1 -1 -2 -2 -2 -1 -2 -3 -2 4 4 3 . . .

. . . -1 -3 -4 -4 -1 -3 -3 -4 -4 2 0 -3 0 -1 -3 -2 -1 3 . . .

. . . 0 -2 0 -1 -1 -1 -1 -1 -2 -2 -2 -1 -2 -3 -2 4 4 3 . . .

. . . -1 -3 -3 -2 -3 -2 -2 -3 -3 -3 -4 -2 -3 -4 8 -1 -2 4 . . .

. . . 4 -2 -1 -2 -1 -1 -1 -1 -2 -2 -2 -1 -2 -3 -1 3 0 3 . . .

Length of protein(l)

PSI-BLAST

. . . N L G A G N S G L N L G H V A L T F . . .

Why Profile-FKNN?

Evolutionary information has been shown to increase the accuracy of secondary structure prediction by many popular methods

An attempt to combine the advantages of incorporating the evolutionary information, fuzzy set theory and nearest neighbor methods

Methods

Calculate profiles using PSI-BLAST The popular Rost and Sander database of 126

representative proteins (<25% sequence Identity)

Find K-Nearest Neighbors Calculate the membership values of the neighbors Calculate the membership values of the current

residue Assign classes Filter the output

Profile Calculation

The profiles of both the query protein and the test protein are calculated using the program PSI-BLAST

Parameters for PSI-BLAST Expectation Value (e) = 0.1

Maximum number of passes (j) = 3

E-value threshold for inclusion in multi-pass model (h) = 5

Default values for the rest of the parameters

K-Nearest Neighbors

For each profile-window in the query protein, the position-weighted absolute distance ‘d’ is calculated from all profile-windows of all proteins in the database.

The profile-windows corresponding to K smallest distances are retained as the K-Nearest Neighbors

1,min,1maxi

Databaseij

Queryij jWjppd

Distance Calculation

. . . 5 -2 -2 -2 -1 -1 -1 0 -2 -2 2 . . .

. . . -1 -3 -2 -2 -3 -2 -2 -3 -3 -3 -3 . . .

. . . 3 -2 -1 -2 -1 -2 -2 2 -2 -2 -2 . . .

. . . 2 -3 -3 -3 -2 -2 -3 -2 -3 1 0 . . .

. . . 0 -2 0 -1 -1 -1 -1 -1 -2 -2 -2 . . .

. . . -1 -3 -4 -4 -1 -3 -3 -4 -4 2 0 . . .

. . . 0 -2 0 -1 -1 -1 -1 -1 -2 -2 -2 . . .

. . . -1 -3 -3 -2 -3 -2 -2 -3 -3 -3 -4. . .

. . . 4 -2 -1 -2 -1 -1 -1 -1 -2 -2 -2 . . .

. . . 0 -1 0 -1 -1 -1 -1 -1 -2 -2 -3 . . .

. . . 0 -3 -1 -2 -3 -2 -3 6 -3 -4 -4 . . .

. . . 0 -3 -4 -4 -2 -3 -3 -3 -3 1 5 . . .

. . . 2 -1 0 -1 -1 -1 -1 -1 -2 -3 -3 . . .

. . . -2 -2 3 6 -4 -1 1 -2 -1 -4 -4 . . .

. . . 2 -3 -3 -3 -2 -2 -3 -2 -3 1 0 . . .

. . . 0 -2 0 -1 -1 -1 -1 -1 -2 -2 -2 . . .

. . . -1 -3 -4 -4 -1 -3 -3 -4 -4 2 0 . . .

. . . 0 -2 0 -1 -1 -1 -1 -1 -2 -2 -2 . . .

. . . -1 -3 -3 -2 -3 -2 -2 -3 -3 -3 -4 . . .

. . . 4 -2 -1 -2 -1 -1 -1 -1 -2 -2 -2 . . .

. . . N L G A G N S G L T F . . .