Chemometrics

Question 1

what is chemometrics

Answer 1

A computationally intensive, multivariate statistical analysis applied to chemical systems or processes

Question 2

why is chemometrics good

Answer 2

reduced large data sets
identifies sample groupings
allows for better visualisation of data
isolates important variables and identifies co-variance

Question 3

EDA (exploratory data analysis)

Answer 3

a pattern recognition technique (identifies groupings and visualises trends)
unsupervised technique
- the basis from which supervised techniques are then followed on from.

Question 4

CA (cluster analysis)

Answer 4

groups samples into clusters based on similarity
- Aglomerative - individual samples –> cluster
- Hierarchal - cluster –> individual sample

Question 5

what is the output from cluster analysis

Answer 5

a dendrogram - simplifies complex data but doesn’t say why the groupings exist only that they do.

Question 6

PCA (principle component analysis)

Answer 6

1. decides features of a data set that are relevant
2. largest variation given highest priority (PC1)
3. if that is not sufficient to describe data then repeat for PC2
4. done until all data is modelled (ideal = as much data is as few PCs)
   data = structure + noise

Question 7

what is a principle component

Answer 7

the linear combinations of the original variables

Question 8

Explained variance

Answer 8

aka scree
the structure = the explained variance and is described by PCs
the EV plot reveals the optimum number of PCs

Question 9

what are scores?

Answer 9

the distance along each PC a sample is from the mean the sample is mapped on to a SCORES PLOT
(can be 3D = increased visualisation)

Question 10

PCA loadings

Answer 10

the coefficients/ weights attached to the variables, so they map the VARIABLES
is the link between the scores plot and the chemistry of the samples

shows how much the original variable contributes to their PC.

Question 11

LDA (linear discriminant analysis)

Answer 11

supervised technique
uses PCs to create classification rules
PCs = ‘discriminant functions’
scores = ‘discriminant values’
tries to minimise variation w/in groups to max separation by prioritising loadings

Question 12

LDA discriminant values

Answer 12

samples in the same group have similar discriminant values (scores) and visa versa

Question 13

caveats

Answer 13

• pre processing is important
• doesn’t compensate for bad data
• potential for cognitive bias
• never a sub for human interpretation