2021 Fiscal Year Annual Research Report
植物-微生物叢相互作用のマルチオミクス階層モデリングとその高速アルゴリズムの開発
| Project/Area Number |
21J21850
|
|---|
| Allocation Type | Single-year Grants |
|---|
| Research Institution | The University of Tokyo |
|---|
Principal Investigator |
Dang Tung 東京大学, 農学生命科学研究科, 特別研究員(DC1)
|
|---|
| Project Period (FY) |
2021-04-28 – 2024-03-31
|
| Keywords | variational inference / stochastic optimization / variable selection / drought irrigation / microbiome |
|---|
| Outline of Annual Research Achievements |
As the substantial increases in the dimensionality of the microbial datasets cause computational burden and poor performance with previous methods, the proposed approach could satisfy the high demands of the microbiome analysis. Our stochastic variational variable selection (SVVS) approach is useful in several aspects. First, SVVS integrated an indicator variable into the framework of the infinite Dirichlet multinomial mixture (DMM) model to identify significant microbiome species (or OTUs) and used stochastic variational inference (SVI) to overcome computational limitations. Thus, the SVVS approach quickly identified the core set of microbial species (or OTUs), considerably improving the performance of the infinite DMM model. In particular, the SVVS method can complete its main tasks in massive microbiome datasets that the previous methods could not perform. Second, a stick-breaking representation was proposed to extend the finite DMM model to an infinite case. This solution treated the total number of clusters as a free parameter of the model, which could help avoid the disadvantages of determining the number of clusters before running the algorithms. Therefore, SVVS could identify the main enterotypes of the healthy human microbiome and detect the important microbiome species that contribute to the variation of the different community compositions.
|
| Current Status of Research Progress |
Current Status of Research Progress
2: Research has progressed on the whole more than it was originally planned.
Reason
Thank to JSPS research funding in 2021, I went to Tottori university in order to collect the microbiome databases in summer and winter. I invested some high-performance computational machines that can analyze the high dimensional microbiome databases. These microbiome databases were used to analyze in my current manuscript (doi: https://doi.org/10.1101/2021.10.04.462986). My current manuscript has been under review process. I have some plans that participate some Japanese and international conferences in order to report my novel methodology.
|
| Strategy for Future Research Activity |
In recent years, several studies have highlighted the substantial role of large-scale analysis in discovering microbiome connections with host metabolism, host genetics in human health, medication, and agroecosystems. An increasing number of multi-omics datasets have been published, such as the integration of metagenomics, metatranscriptomics, metaproteomics, whole-genome sequencing, and whole-transcriptome sequencing of the TCGA cancer microbiome. In the future, we plan to extend the SVVS approach to a comprehensive analysis of multi-omics datasets. The main framework of the SVVS method can be developed for the other Bayesian mixture models such as beta-mixture models for microarray gene expression datasets, and multinomial mixture model for ChIP-exo sequencing data. Therefore, this method provides to new opportunities for discovering the significant associations of microbes with specific nutrients and medication or the important interactions between plants, microbes, and soils. Moreover, the dimensionality of databases significantly increased when other datasets were combined with microbiome data. We considered the integration of parallel computation strategies that can partition the data both horizontally (over sample) and vertically (over features) to speed up the computational processes.
|
-
-
[Journal Article] Stochastic variational variable selection for high-dimensional microbiome data2021
Author(s)
Dang, Tung and Kumaishi, Kie and Usui, Erika and Kobori, Shungo and Sato, Takumi and Ichihashi, Yasunori and Yusuke, Toda and Yamasaki, Yuji and Tsujimoto, Hisashi and Iwata, Hiroyoshi
-
Journal Title
bioRxiv
Volume: 1
Pages: 1-19
DOI
Open Access
