Lecture 15 - Gene Expression Statistics

Question 1

what is accuracy

Answer 1

how close something is to the real answer

Question 2

what is precision

Answer 2

how close your replicates are (consistency)

Question 3

what is good precision characterized by

Answer 3

reproducible results

Question 4

what is good accuracy characterized by

Answer 4

measurements that correspond to an independently known result

Question 5

what is the main goal of data preprocessing

Answer 5

• removing systematic bias and variation in the data due to non-biological factors
• preserving the variate in gene expression that occurs because of biologically relevant changes in transcription

Question 6

what are two methods of global normalization

Answer 6

• Trimmed Means of Means (TMM)
• use housekeeping genes

Question 7

what is Trimmed Mean of Means (TMM)

Answer 7

• remove genes with big difference (log-fold change) between samples, or very high or low expression
• adjust sample to a common mean after removing trimmed genes

Question 8

why is log fold-change used for TMM [log(A/B)]

Answer 8

• proportional changes are generally more important that absolute changes in biological data

Question 9

what are housekeeping genes

Answer 9

genes that are typically invariant

Question 10

what is a plot commonly used to represent gene expression values from two samples

Answer 10

scatter plots

Question 11

how do log transformations change a scatter plot of gene expression

Answer 11

it results in a more even distribution of data points and the variance is more uniform

Question 12

what is inferential statistics used for

Answer 12

used to make inferences about a population from a sample

Question 13

what type of statistics is hypothesis testing a form of

Answer 13

inferential statistics

Question 14

what is the null hypothesis for gene expression

Answer 14

there is no difference in gene expression between samples

Question 15

what is the alternate hypothesis for gene expression

Answer 15

there is a difference in expression between samples

Question 16

how do we determine whether or not to reject the null hypothesis

Answer 16

with statistical tests

Question 17

what is type I error of hypothesis testing

Answer 17

a false positive; the null hypothesis is rejected but it is actually true

Question 18

what is type II error in hypothesis testing

Answer 18

a false negative; we fail to reject the null hypothesis but it is actually false

Question 19

what can p-values be used to estimate in terms of hypothesis testing

Answer 19

type I errors, false positives

Question 20

what does it mean if the p-value is less than a threshold α

Answer 20

the chance of a false positive is less than α

Question 21

what is a common form of hypothesis testing

Answer 21

the t-test

Question 22

what is the issue with p-values and multiple hypothesis testing

Answer 22

you would expect to see 5% identified at the p < 0.05 level by chance alone

Question 23

what is the Bonferroni correction

Answer 23

the level of statistical significance divided by the number of measurements

Question 24

what does the new threshold created by the Bonferroni correction mean

Answer 24

the chance of any false positive being present is < 0.05

Question 25

what is the issue with the Bonferroni correction

Answer 25

it will reject many true positive cases

Question 26

what is the false discovery rate (FDR)

Answer 26

Expected(# false positives / # total positives)

Question 27

how do we estimate the number of expected false positives for any p-value threshold

Answer 27

approximate the background distribution of negatives based on the null hypothesis

Question 28

what does a greater difference from the background distribution imply for the p-value

Answer 28

it is smaller

Question 29

how do you calculate q, the false positive rate

Answer 29

expected number of false positives / total number of positives

Question 30

what do we want to find for FDR

Answer 30

the largest i such that p_(i) <= (i/m) * q

Question 31

what is descriptive statistics

Answer 31

finding meaningful patterns in data

Question 32

what are two commonly used distance metrics

Answer 32

Euclidian Distance and Pearson Correlation Coefficient

Question 33

how is Euclidian distance calculated

Answer 33

d_xy = the square root of the sum of the differences squared for 3D: [(x1-y1)^2+(x2-y2)^2+(x3-y3)^2]^0.5

Question 34

what is the calculation for the Pearson Correlation Coefficient (r)

Answer 34

* search it up

Question 35

what does an r value of 0 (Pearson Correlation Coefficient) indicate

Answer 35

X and Y are uncorrelated

Question 36

how can the Pearson Correlation Coefficient be translated to a rough distance measure

Answer 36

D = 1 - r

Question 37

what are the two types of hierarchical clustering

Answer 37

agglomerative and divisive

Question 38

what is agglomerative hierarchical clustering

Answer 38

building up the branches of a tree, beginning with the two most closely related objects

Question 39

what is divisive hierarchical clustering

Answer 39

building the tree by finding the most dissimilar group first

Question 40

what is the distance between clusters in UPGMA

Answer 40

the average of all distances between cluster elements

Question 41

what are three ways distances between clusters are defined

Answer 41

- single linkage clustering - complete linkage clustering - centroid clustering

Question 42

what is single linkage clustering

Answer 42

the minimum distance between a point in X and a point in Y

Question 43

what is complete linkage clustering

Answer 43

the maximum distance between a point in X and a point in Y

Question 44

what is centroid clustering

Answer 44

the distance between the average within clusters (centroids)

Question 45

how do you get a correlation coefficient from cluster distances

Answer 45

use the log of expression ratios

Question 46

what range is the distance between clusters in

Answer 46

[0,2]

Question 47

what are two non-hierarchical clustering methods

Answer 47

- k-means clustering - self-organizing maps (SOM)

Question 48

how is k-means clustering done

Answer 48

- randomly pick k points, called centroids - assign all other points to the nearest centroid - recalculate centroids to minimize the sum of squared distances within clusters - iteratively reassign points until convergence

Question 49

how do self-organizing maps differ from k-means clustering

Answer 49

- initial assignment is not random, you pick the starting points - you start with an initial geometry of 'nodes' (e.g. a 3x2 grid)

Question 50

what are self-organizing maps useful for

Answer 50

mapping time-series data