Predicting peptide bond conformation using feature selection and the Naive Bayes approach

Send This

Download

Comment

Favourite

more

Rating :

Rate It:

Embed :

Post a comment

Post Comment on Twitter Post Comment on SlideWorld

Comments:

Subscribe to follow-up comments

Notes

Hide Notes

Slide 1 :

Predicting peptide bond conformation using feature selection and the Naïve Bayes approach K.P. Exarchos1, T.P. Exarchos1, C. Papaloukas1,2, A.N. Troganis2 and D.I. Fotiadis1,3 1 Unit of Medical Technology and Intelligent Information Systems, Department of Computer Science, University of Ioannina, Ioannina, Greece 2Dept. of Biological Applications and Technology, University of Ioannina, Ioannina, Greece 3 Biomedical Research Institute, Ioannina, Greece

Slide 2 :

Presentation outline Objectives Introduction State of the art Materials and methods Results Conclusions

Slide 3 :

Objectives Important role in the protein structure and function Prediction of peptide bond conformation between any two amino acids Utilization of a large and informative feature vector Contribution of each feature towards the prediction of the peptide bond formation

Slide 4 :

a-carbon(Ca) Amino group Carboxyl group Side chain Amino acids – peptide bond formation

Slide 5 :

Peptide bond Peptide bonds can exist in one of two conformational isomers, with the two a-carbons either cis or trans to each other The Ca(i)-Ca(i+1) distance in cis conformation is nearly 1Å shorter than in the trans conformation The majority of the peptide bond conformations in protein structures is found to be trans In the case of X-Pro peptide groups the situation is different since the sterical differences between cis and trans conformations are minimal

Slide 6 :

Cis/trans isomerization (1/2) Only 0.03% of the X-nonPro and 5.2% of the X-Pro peptide bonds are in cis conformation1 The cis/trans ratio depends on the primary amino acid sequence2 There is strong correlation between the resolution of the protein structure and the cis conformation content1 2Grathwohl et al. Biopolymers, 1981 1Weiss et al. Nat. Struct. Biol., 1998

Slide 7 :

Cis/trans isomerization (2/2) The occurrence of cis peptide bonds may be affected by the secondary structure element they belong1 There is a slight preference of cis isomers to surface accessible areas1 The physicochemical properties have been proven to contribute considerably in the discrimination between the peptide bond isomers2 1Pahlke et al. BMC Struct. Biol., 2005 2Frommel et al. FEBS. Lett., 1990

Slide 8 :

Significance of cis/trans isomerization (1/2) Cis peptide bonds, especially the ones between non-proline residues, are frequently located1: Near the active site of proteins In regions related to the function of the protein molecule Cis prolyl residues are more often conserved than the surrounding amino acids and the whole protein 2 The function of some proteins depends on a certain cis peptide bond3,4 2Lorenzen et al. Proteins, 2005 4Brauer et al. Biochemistry, 2002 1Jabs et al. J. Mol. Biol., 1999 3Lummis et al. Nature, 2005

Slide 9 :

Significance of cis/trans isomerization (2/2) The possible biological role of cis/trans isomerization in protein splicing and energy reservoir is still a matter of debate1 The isomerization of cis prolyl peptide bonds is catalyzed by the peptidyl prolyl isomerases , which are also implicated in2: Cell signalling and replication The induction of severe diseases such as cancer, AIDS, Alzheimer’s disease and other neurodegenerative disorders3 Several cis peptide bonds play a vital role in the4: Final structure and function of proteins Folding and stability of many protein molecules 2Jabs et al. J. Mol. Biol., 1999 1Fischer et al. Rev. Physiol. Biochem. Pharmacol., 2003 Hence, accurate discrimination between the cis/trans conformations, will greatly contribute towards: Reliable prediction of protein structure and function Accurate design of protein molecules 3Dugave et al. Chem Rev., 2003 4Pal et al. J. Mol. Biol., 1999

Slide 10 :

State of the art (1/2) Pattern extraction, based on the amino acid physicochemical properties, for the proline cis/trans isomerization prediction1 Small dataset, prediction only for X-Pro, deficient feature vector SVM with polynomial kernel and single sequence information, coded in binary form for the peptidyl prolyl cis/trans isomerization prediction2 Prediction only for X-Pro, deficient feature vector 1Frömmel et al. FEBS Lett., 1990 2Wang et al. J. Peptide Res., 2004

Slide 11 :

An extension of Chou-Fasman parameters, and the secondary structure of amino acid triplets, for the prediction of cis/trans isomerization between any two amino acids1 Inadequate results, small number of attributes, narrow sliding window SVM with RBF kernel and multiple sequence alignment coupled with the secondary structure, for the proline cis/trans isomerization prediction2 Prediction only for Prolines, missing attributes State of the art (2/2) 1Pahlke et al. Bioinformatics, 2005 2Song et al. BMC Bioinformatics, 2006

Slide 12 :

Proposed method Feature extraction Feature selection FASTA sequence Peptide bond classification

Slide 13 :

Feature extractionPSSM PSI-BLAST (Position Specific Iterated Blast)1 : multiple sequence alignment profiles in the form of position specific scoring matrices (PSSM) Distant evolutionary relationships between proteins ?-value=10-3, 3 iterations, nr database PSSM: frequencies of each amino acid in each position of protein sequence 1Altschul et al. Nucleic Acids Res., 1997

Slide 14 :

Feature extraction Secondary structure Predicted secondary structure information PSIPRED1, reliability indices for all three secondary states for each residue in the query sequence PSSMs as input and two feed forward neural networks for prediction 1Jones J. Mol. Biol., 1999

Slide 15 :

Feature extractionSolvent accessibility Predicted relative solvent accessibility RVP-net1 : predictions about the relative solvent accessibility of each residue Real valued predictions of accessible surface area for each amino acid 1Ahmad et al. Bioinformatics, 2003

Slide 16 :

Feature extractionPhysicochemical properties We also employed six properties for each amino acid1: Volume Hydrophobicity Polarity Charge Aromatic/aliphatic character For volume, hydrophobicity and polarity, real valued indices were used 1Grantham Science, 1974

Slide 17 :

Feature selection Approaches: Filter approaches are independent of the classification task and are based on certain metrics like correlation to evaluate features or subsets of features Wrapper approaches use the target learning algorithm as a black box to estimate the worth of attribute subsets by measuring accuracy estimates Feature wrappers often achieve better results than filters due to the fact that they are tuned to the target data mining algorithm1 1Tan et al. Introduction to data mining, 2006

Slide 18 :

Peptide bond classificationNaïve Bayes Naïve Bayes Given a test instance with attributes from the set we aim to predict the class it belongs Using the Bayes theorem we can compute the posterior probability for each class C We choose the class that maximizes the above quantity

Slide 19 :

FASTA sequence Physicochemical properties (w*6) PSSM (w*20) Secondary structure (w*3) Solvent accessibility (w*1) PSI-BLAST PSIPRED RVP-net 931 features Feature extraction Feature selection Feature selection search Induction algorithm Feature evaluation Feature set Naïve Bayes classifier Peptide bond classification

Slide 20 :

Dataset The dataset used includes 3050 protein sequences obtained from the Protein Data Bank1 (PDB) Sequence identity<25% R-factor <0.25 Resolution <2.0Å The annotation of the dataset was performed using Volume Area Dihedral Angle Reporter (VADAR)2 1Berman et al. Nucleic Acids Res., 2000 2Willard et al. Nucleic Acids Res., 2003

Slide 21 :

Class imbalance problem Sample-based approaches: Oversampling Undersampling Hybrid We performed full undersampling 10 times thus creating 10 fully balanced datasets, each containing all cis peptide bonds and equal number of randomly selected trans conformations In each dataset we performed 10-fold cross validation and averaged the results in order to obtain a reliable error estimate

Slide 22 :

Results (1/2) Results obtained with and without the use of the wrapper feature selection step. Mean values are shown for each evaluation metric over the 10 datasets

Slide 23 :

Results (2/2) Results of each attribute in the prediction performance

Slide 24 :

Comparison

Slide 25 :

Conclusions (1/2) Prediction both for X-Pro and X-nonPro peptide groups Utilization of a large and informative feature vector, incorporating many of the features affecting the peptide bond conformation Powerful and efficient feature selection in order to remove redundant and irrelevant features Employment of a simple and fast classification algorithm such as the Naïve Bayes

Slide 26 :

Conclusions (2/2) Analysis of the contribution of each attribute in the classification performance PSSMs and the physicochemical properties: highly discriminatory Secondary structure and the solvent accessibility: very poor discriminative potential

Management of Conges...

An Evidence-based Ap...

A Hands On Approach ...

Revision of a Medica...

Using Patient Satisf...

Alternative Approach...

Free Powerpoint Templates

	kexarcho@cc.uoi.gr 5 Years ago.



778 Views, 0 favourite
PowerPoint Presentation on Predicting peptide bond conformation using feature selection and the Nai more PowerPoint Presentation on Predicting peptide bond conformation using feature selection and the Naive Bayes approach or PowerPoint Presentation on cis/trans isomerization, peptide bond conformation, feature selection, pssm, Naive Bayes less
More By User Flag as inappropriate Select your reason for flagging this presentation as inappropriate. If needed, use the feedback form to let us know more details. Cancel