Lecture 6: Chemometrics

Question 1

What is chemometrics?

Answer 1

Computationally intensive, multivariate (many variables) statistical analysis, applied to chemical systems or processes.

Question 2

What can chemometrics do?

Answer 2

• Reduce complex datasets
• Identify and quantify sample groupings
• Optimise experimental parameters
• Isolate important variables and identify covariance.
• Provide reproducible measures of data
• Allows for better visualisation of data
• Can isolate the important variables
• Removes subjectivity, but it isn’t wholly non-subjective as humans need to interpret the data.

Question 3

When did chemometrics first get used in journals?

Answer 3

it wasn’t used in journals util 1980’s due to skeptism.

Question 4

Where is chemometrics routinely used?

Answer 4

Routinely used in industry for process optimisation and quality control, e.g. food and pharmaceutical
to maximise output and quality with minimal cost.

Question 5

When did chemometrics start getting used in forensic science?

Answer 5

2009

Question 6

What highlighted the need for chemometrics?

Answer 6

NAS reported the need for a ‘statistical framework’

Question 7

What are the categories of chemometrics?

Answer 7

• Design of experiments
• Exploratory data analysis
• Regression
• Classification

Question 8

What is the purpose of DOE?

Answer 8

• Make experiments more effective
• Achieve maximum data
• Improve efficiency, quality and reproducibility

Question 9

What is regression analysis?

Answer 9

• Chemometric version of a calibration curve
• Based on a y = mx + c linear relationship for multiple variables
• Maps the effect of multiple independent variables (predictors) upon dependent variable (response)
• Allows prediction of quantitative sample properties

Question 10

What is dimensionality reduction

Answer 10

Reducing data that has many variables into just a few measures called principle components.

Question 11

What form of chemometrics used dimensionality reduction?

Answer 11

Exploratory Data Analysis (EDA)

Question 12

What are the main features of EDA?

Answer 12

• Dimensionality reduction
• Pattern recognition technique
• Visualise trends that may have gone unnoticed.
• Determination of sample similarity in complex data
• Unsupervised

Question 13

What does pattern recognising technique do?

Answer 13

It identifies groupings within data.

Question 14

What is an unsupervised technique?

Answer 14

Exploring the data without any prior assumptions or knowledge of the samples

Question 15

What is a supervised technique?

Answer 15

Supervised is when you’re building classification rules once you know the groupings of the sample and how they compare.

Question 16

What are the most commonly used EDA techniques?

Answer 16

• Cluster Analysis (CA)
• Principle Component Analysis (PCA)

Question 17

What type of technique is cluster analysis?

Answer 17

• Unsupervised
• Form of EDA

Question 18

What does cluster analysis do?

Answer 18

Samples grouped into clusters based on calculated distance (similarity)

Question 19

What are the two types of cluster analysis?

Answer 19

Agglomerative
Hierarchical

Question 20

What is the agglomerative technique?

Answer 20

• This is when you go from individual samples to clusters
• The samples become more dissimilar
• Ones grouped close to 0 are the most similar

Question 21

What is hierachical?

Answer 21

This is when you go from clusters to individual samples

Question 22

How is data from cluster analysis presented?

Answer 22

On a dendrogram

Question 23

What do analysts need to decide for cluster analysis?

Answer 23

They need to decide the stopping rules to determine the number of clusters arbitrarily

Question 23

What is the problem with cluster analysis?

Answer 23

it can only tell you there are groupings and not why.

Question 24

Why isn’t cluster analysis entirely non-subjective?

Answer 24

Because the analyst decides the stopping point

Question 25

What technique is principle component analysis?

Answer 25

• Unsupervised technique
• Form of EDA
• Dimensionality reduction technique

Question 26

What does PCA do?

Answer 26

Assesses all variables within a dataset (eg spectrum) and deecides which are relevant, then determines which variables are correlated

Question 27

What does the algorithm do in PCA when it finds a correlation?

Answer 27

Wherever the algorithm finds correlation it defines this as a Principal Component (PC)

Question 28

What PC is given the highest priority?

Answer 28

The PC that describes the largest variation between samples will be given the highest priority, i.e. PC1