應用多重關聯式動態分群演算法於次世代轉錄體定序資料分群之研究

孫榕辰 ； 陳榮銘 ； 胡若梅

《TANET2017 臺灣網際網路研討會》 2017年 (2017/10) Pp. 7-12

https://doi.org/10.6728/TANET.201710.0002

摘要

本研究探討在具全域關聯性權重多重關聯式動態分群演算法(MRDCA_RWG)相異度矩陣加上設限條件，並應用於次世代轉錄體定序(RNA-seq)資料的分群，研究在關聯性相異度矩陣加上適當設限權重(稱為改良式MRDCA_RWG)是否能有效提升植基於多重關聯式動態分群演算法的分群劃分成效。概念是若是能夠於各種不同特徵關聯式相異度矩陣加上適當的設限，應有助於導引演算法收斂至一組更合適的區域最佳分群劃分解。本研究並使用過去學者發表的玉蜀黍葉的轉錄體資料進行驗證與比較，分群成效是使用一般常用的外部量化分群評估指標NMI (normalized mutual information)。實驗結果顯示本研究提出的改良式MRDCA_RWG比MRDCA_RWG擁有更優的分群成效，顯示改良式MRDCA_RWG能有效提升植基於多重關聯式動態分群演算法的分群劃分成效。

關鍵字

多重關聯式動態分群演算法 ； 關聯性權重 ； 轉錄體定序基因表現 ； 機率模型 ； 基因語意相似度

並列摘要

This study considered the problem of constraint setting for the dissimilarity matrices in multiple relational dynamic clustering algorithm with relevance weight of each dissimilarity matrix estimated globally (MRDCA_RWG). Inspired by the idea that imposing suitable constraints on the dissimilarity matrices may lead the clustering partition into a better local optimal one, a modified MRDCA_RWG with weighted geometric mean on the dissimilarity matrices was investigated with applications to clustering of RNA-seq data. The effectiveness of MRDCA_RWG and modified MRDCA_RWG was validated by applying them to real-life maize RNA-seq dataset obtained from past related studies. The clustering results were analyzed and compared using a well-known external criterion for clustering evaluation, the normalized mutual information (NMI). Experimental results indicated that the proposed modified MRDCA_RWG can produce more valid clustering results than the MRDCA_RWG.

並列關鍵字

multiple relational dynamic clustering algorithm ； relevance weights ； RNA-seq gene expression ； probability model ； gene semantic similarity