350 likes | 362 Views
Lilikoi is a new R package specializing in personalized pathway measurement and classification prediction models in metabolomics research. It offers features like Metabolites-Pathway mapping, Pathway dysregulation score deriving, Feature selection, and Classification & performance measurement using machine learning algorithms. Lilikoi allows for the identification and prediction of biomarkers based on personalized pathway analysis.
E N D
Lilikoi: Metabolomics Personalized Pathway Analysis Tool Fadhl Alakwaa, Ph.D Department of Computational Medicine and Bioinformatics University of Michigan https://github.com/FADHLyemen/
Metabolomics research* *https://www.sciencedirect.com/science/article/pii/S1740674915000074
Why metabolomics? “The deep insight into the actual phenotype of any biological system is metabolomics’ advantage over other -omicstechnologies” * *https://www.metanomics-health.com/en/why-metabolomics.html
Metabolomic pathways analysis tools MetPA IMPaLA MPEA The classical approaches use statistical models (e.g. the hypergeometric test) to calculate the probability of observing the actual number of differentially metabolites in a given pathway by chance.
Metabolomics pathways analysis tools: why Lilikoi? MetPA IMPaLA MPEA None of these metabolomics pathway-based tools employ pathways as features for downstream biomarker modeling.
Metabolomics pathways analysis tools. why Lilikoi? MetPA IMPaLA MPEA none of these pathway-based methods transform the metabolite-sample matrix into pathway-sample matrix, to entail pathway representation at individual sample level (or personalized level).
Metabolomics pathways analysis tools. why Lilikoi? MetPA IMPaLA MPEA We can not used these tools to construct pathways features for downstream analysis such as classification and clustering
Lilikoi Lilikoi is a new R package which specializes in personalized pathway measurement and classification prediction models. https://github.com/lanagarmire/lilikoi
Metabolites-pathway Metabolites
Module 2. Pathway dysregulation score deriving* *Drier et. al (2013) PNAS
Module 3. Feature selection • Two major feature selection algorithms Implemented in Rweka: • information gain (mutual information) • gain ratio
Module 4. Classification & performance measurement • Machine learning algorithms (LDA, SVM, RF, RPART, PAM, LOG ) • Parameters auto tuning • nfold-cross classification • Performance measurement(AUC, F1, SEND, SPEC) • Importance ranking
Example dataset • For demonstration, we present a metabolomics data set from the City of Hope Hospital (COH). This dataset is composed of 207 samples from plasma (126 cases and 81 controls).
Lilikoi run options: • Shinny • http://lilikoi.garmiregroup.org
Lilikoi run options: • My binder • https://mybinder.org/v2/gh/FADHLyemen/lilikoi_Fadhl/master
Lilikoi run options: • Docker • docker pull fadhlyemen/lilikoi • docker run -d --rm -ti -p 5001:8888 fadhlyemen/lilikoi start-notebook.sh --NotebookApp.token='‘
Lilikoi run options: • install.packages("lilikoi") • # Or for the latest dev version: • devtools::install_github("lanagarmire/lilikoi")
Lilikoi run • https://mybinder.org/v2/gh/FADHLyemen/lilikoi_Fadhl/master • https://mybinder.org/v2/gh/FADHLyemen/lilikoi_Fadhl/master
Lilikoi run • https://mybinder.org/v2/gh/FADHLyemen/lilikoi_Fadhl/master
Lilikoirun: metabolomics mapping • Lilikoi allows the user to input any kind of metabolite IDs, their synonyms, KEGG IDs, HMDB IDs or PubChem IDs. • Lilikoi embeds comprehensive databases including over 18,000 metabolites and 100,000 synonyms. • If metabolite was not found in our system, fuzzy matching was implemented, by calculating the string similarity score of the input metabolite name with those in the databases.
Future work • Update the database to increase metabolites mapping • Automl
Summary • Current pathway-based methods in the metabolomics field are not personalized and they are merely used for graphical mapping and enrichment analysis. • None of these metabolomics pathway-based tools employ pathways as features for downstream biomarker modeling. • Lilikoi addresses all these issues with personalized pathway deregulation measurements (PDS scores) and offers a standardized classification model for biomarker prediction.
Acknowledgements • Financial support: • K01ES025434 awarded by NIEHS through funds provided by the trans-NIH Big Data to Knowledge (BD2K) initiative (www.bd2k.nih.gov) • P20 COBRE GM103457 awarded by NIH/NIGMS • R01 LM012373 awarded by NLM • Hawaii Community Foundation Medical Research Grant 14ADVC-64566 • Garmire lab members