東華大學圖書館 |

Language: English

Help

回圖書館首頁

手機版館藏查詢

Back

Switch To: Labeled | MARC Mode | ISBD

A machine learning approach to prote...

Pollastri, Gianluca.

Linked to FindBook

Google Book

Amazon

博客來

A machine learning approach to protein structure prediction.

Record Type:	Language materials, printed : Monograph/item
Title/Author:	A machine learning approach to protein structure prediction./
Author:	Pollastri, Gianluca.
Description:	166 p.
Notes:	Chair: Pierre F. Baldi.
Contained By:	Dissertation Abstracts International64-01B.
Subject:	Biology, Genetics. -
Online resource:	http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=3078784
ISBN:	0493999000

A machine learning approach to protein structure prediction.
Pollastri, Gianluca.

A machine learning approach to protein structure prediction. - 166 p.

Chair: Pierre F. Baldi.

Thesis (Ph.D.)--University of California, Irvine, 2003.

This Thesis focuses on the development of machine learning algorithms for structured data and their application to the prediction of protein structure from its primary sequence.

ISBN: 0493999000Subjects--Topical Terms:

1017730
Biology, Genetics.

A machine learning approach to protein structure prediction.
LDR:03722nam 2200361 a 45 001 928311
005 20110426
008 110426s2003 eng d
020 $a 0493999000
035 $a (UnM)AAI3078784
035 $a AAI3078784
040 $a UnM $c UnM
100 1 $a Pollastri, Gianluca. $3 1251772
245 1 0 $a A machine learning approach to protein structure prediction.
300 $a 166 p.
500 $a Chair: Pierre F. Baldi.
500 $a Source: Dissertation Abstracts International, Volume: 64-01, Section: B, page: 0290.
502 $a Thesis (Ph.D.)--University of California, Irvine, 2003.
520 $a This Thesis focuses on the development of machine learning algorithms for structured data and their application to the prediction of protein structure from its primary sequence.
520 $a We then derive a modified version of the architecture to deal with outputs of quadratic size in the length of the input. We apply the model to the prediction of β-sheet partnerships between residues in proteins.
520 $a We first develop a novel family of recurrent neural networks for the translation of one-dimensional data of variable length. These architectures introduce non-causal bidirectional dynamics to capture both upstream and downstream information in sequences. We derive an inference algorithm for the new architectures, and we formulate the learning algorithm as a maximum likelihood estimation. We apply the architecture to the prediction of protein secondary structure obtaining state-of-the-art results, and we study its effective capability of capturing long-range information.
520 $a Thirdly we build two secondary structure prediction systems using ensembles of bidirectional recurrent neural networks and inputs containing evolutionary information in the form of multiple alignments of homologue sequences: (1) one for secondary structure classification into three classes; (2) one for secondary structure classification into eight classes. We assess the results of the two systems of three different test sets on which the three-class predictor achieves a performance at or above the state of the art. The systems are implemented as web servers and at the moment of writing have served over 45,000 queries from all over the world.
520 $a We build two more prediction systems, one for the classification of the number of stabilizing contacts in proteins, the other for the prediction of residue solvent accessibility. The systems show performances above the state of the art in a three-fold cross-validation experiment. Furthermore, we prove the capability of bidirectional recurrent neural networks to capture distant information in the case of contact number prediction. The two systems are also implemented as web servers.
520 $a Finally, we extend the concept of bidirectional recurrent neural networks to N-dimensional translations. We formalize inference and learning algorithms for the 2-dimensional case, where contextual information is propagated laterally through four hidden planes, one for each cardinal corner. We show that these architectures yield protein contact map predictors that outperform previous methods. Contacts between non-neighboring amino acids are classified with a precision above 50% for all tested cutoffs, close to levels considered compatible with accurate determination of protein three-dimensional structure.
590 $a School code: 0030.
650 4 $a Biology, Genetics. $3 1017730
650 4 $a Biology, Molecular. $3 1017719
650 4 $a Biophysics, General. $3 1019105
650 4 $a Computer Science. $3 626642
690 $a 0307
690 $a 0369
690 $a 0786
690 $a 0984
710 2 0 $a University of California, Irvine. $3 705821
773 0 $t Dissertation Abstracts International $g 64-01B.
790 $a 0030
790 1 0 $a Baldi, Pierre F., $e advisor
791 $a Ph.D.
792 $a 2003
856 4 0 $u http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=3078784