Using R to Make Sense of NMR Datasets

Transcription

1 Using R to Make Sense of NMR Datasets Prof. Bryan A. Hanson Dept. of Chemistry & Biochemistry DePauw University, Greencastle Indiana Presentation available at github.com/bryanhanson/panic2017 Additional references & resources on last slide

2 What Exactly is R? R is a free software environment for statistical computing and graphics The Ecosystem: Base R via R-Core Add-on packages from many authors Comprehensive R Archival Network (aka CRAN) (>10,000 packages) Bioconductor (>1,300 packages) Unofficial repositories: Github, Gitlab, SourceForge etc. Support forums User guides galore!

3 The Ecosystem: Support Resources Official Documentation Focused, Topical Task Views R-Bloggers: over 600 R-oriented bloggers Stack Overflow: over 160K questions on use of R Hundreds of Intro to R documents on the web Dozens of free R books on the web Many packages have a vignette or user guide. More resources on last slide.

4 Features of R Written by statisticians... a rather unlikely linguistic cocktail... 1 Cross-Platform: Windows, Linux, Mac OS Infrastructure: ready integration, interactive options Interfaces to many other languages, programs SAS, SPSS, python, JavaScript, MATLAB, C++ etc. Several ways of running in parallel, using multiple cores Command line, or several GUI options 1 Structure of the R Language

5 R is Open Source Free! Transparent: All code readily available for inspection Given enough eyeballs, all bugs are shallow Linus Torvalds Many parts of the ecosystem are community driven Open source means everyone can see my stupid mistakes. Version control means everyone can see every stupid mistake I ve ever made. Karl Broman

6 Do People Use R? 2 2 Bob Muenchen r4stats.com/articles/popularity/

7 Who Uses R? AirBnB Zillow 3 Etsy NYT Twitter Facebook 3 Data Science at Zillow

8 User Contributed Packages 4 4 Script by Gergely Daróczi

9 Reproducible Research with R Automation of Workflow: data analysis code + explanatory text figures + tables + text = report Many resources for reproducible research Several possible input formats Typical output formats are pdf files and web pages This presentation written with L A TEX and R via the knitr package.

10 What is ChemoSpec? ChemoSpec = Chemometrics + Spectroscopy Tools for exploratory data analysis No attempt to duplicate functions available on the spectrometer

11 ChemoSpec: Design Goals User friendly design Helpful error messages Reliable results High quality plots Consistent plot appearance Provide access to a wide range of chemometric operations Extensibility Developed with metabolomics and IR, NMR & Raman in mind

12 What Can ChemoSpec Do? Data Cleaning & Prep Import data Remove samples Drop frequency ranges Baseline correction Signal alignment Normalization Savitzky-Golay filters Exploratory Data Analysis Plotting & surveying Hierarchical cluster analysis (HCA) Principal component analysis (PCA) PCA diagnostics Score & loading plots ANOVA-PCA Empirical clustering

13 Demonstration Data Set: Saw Palmetto Caps Retail samples of Serenoa repens gel caps 500 MHz 1 H NMR in CDCl 3 4 samples were pure according to the label 10 samples have another oil present per label 2 outliers: olive oil, and evening primrose oil Serenoa repens extracts mainly fatty acids Outliers mainly triglycerides

14 Representative 1 H NMR Spectra peak height 0.0e e e+07 TJ_OO ET_pSrE ppm

15 Where is the Variation in the 1 H NMR Spectra? Centered Spectra 2.0e e e e ppm

16 Hierarchical Clustering 0e+00 4e+07 8e+07 SV_EPO TJ_OO NP_adSrE NR_pSrE ET_pSrE MDB_pSrE LF_adSrE SN_adSrE clustering method: complete NA_pSrE Nat_adSrE GNC_adSrE NW_adSrE NSI_adSrE distance method: euclidean TD_adSrE CVS_adSrE Sol_adSrE Key adsre EPO OO psre

17 Principal Component Analysis: Scree Plot 2e+07 2e+07 6e+07 component scores % 81% 91% 95% 97% 98% 99% 99% 100% 100% centered/noscale/classical cumulative percent variance shown to right of PC

18 Principal Component Analysis: Score Plot PC2 score (20%) 4e+07 0e+00 4e+07 centered/noscale/classical classic ellipses by group Key adsre EPO OO psre 4e+07 2e+07 0e+00 2e+07 4e+07 6e+07 PC1 score (62%)

19 Principal Component Analysis: Loadings Plot Reference Spectrum PC 1 Loadings ppm centered/noscale/classical

20 Principal Component Analysis: S Plot 1e+14 5e+13 0e covariance correlation centered/noscale/classical

21 Acknowledgements Thanks for your attention! Kristie Adams for the invite Sabbatical Support, DePauw University

22 Additional References & Resources R Project Home Page Selected Topical Task Views Chemometrics & Computational Physics Clinical Trials Experimental Design Pharmacokinetics Machine Learning Reproducible Research Bioconductor Home Page