Class 9: Linear Models & Enrichment

Methodology of Scientific Research

Andrés Aravena, PhD

9 Jun 2021

When to use Students’ t

In the cases we see here, we assume that the underlying distribution is Normal

(that is usually the case for averages)

We need to know the population mean and standard deviation

Students’ t when we don’t know population standard deviation

When we have a sample of size 𝑛, we calculate

  • sample mean
  • sample standard deviation

and we use them to approximate the population values

Then we use Student’s t distribution

Example: gene expression

Genotype Stress logExpr
1 WT Low 4.439524
2 WT Low 4.769823
3 WT Low 6.558708
4 WT High 7.070508
5 WT High 7.129288
6 WT High 8.715065
7 KO Low 6.460916
8 KO Low 4.734939
9 KO Low 5.313147
10 KO High 7.554338
11 KO High 9.224082
12 KO High 8.359814

Visually

Using a Linear Model

We want to find the change in gene expression due to genotype and stress condition

A common approach is to model gene expression as \[\text{Gene expression} = β_0 + β_1⋅\text{Genotype} + β_2⋅\text{Stress}\]

Here Genotype and Stress are either 0 or 1, depending on the case, and we want to find \(β_0, β_1,\) and \(β_2\) (the coefficients)

Model Matrix

(Intercept) GenotypeKO StressHigh
1 1 0 0
2 1 0 0
3 1 0 0
4 1 0 1
5 1 0 1
6 1 0 1
7 1 1 0
8 1 1 0
9 1 1 0
10 1 1 1
11 1 1 1
12 1 1 1

Fitting the model in R

Most people use the R package limma for gene expression

Here we show a simple version, with one single gene

model <- lm(logExpr ~ Genotype + Stress)
model

Call:
lm(formula = logExpr ~ Genotype + Stress)

Coefficients:
(Intercept)   GenotypeKO   StressHigh  
     5.1325       0.4941       2.6293  

Estimated coefficients are the averages from the sample

Looking at the fitted model

summary(model)

Call:
lm(formula = logExpr ~ Genotype + Stress)

Residuals:
    Min      1Q  Median      3Q     Max 
-0.8916 -0.6917 -0.3380  0.8641  1.4262 

Coefficients:
            Estimate Std. Error t value Pr(>|t|)    
(Intercept)   5.1325     0.4553  11.273 1.31e-06 ***
GenotypeKO    0.4941     0.5257   0.940 0.371872    
StressHigh    2.6293     0.5257   5.001 0.000738 ***
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

Residual standard error: 0.9106 on 9 degrees of freedom
Multiple R-squared:  0.7421,    Adjusted R-squared:  0.6848 
F-statistic: 12.95 on 2 and 9 DF,  p-value: 0.002247

Estimated coefficients follow a t distribution

Confidence interval for the coefficients

Estimated coefficients follow a t distribution

We have 12 values, and we evaluate 3 coefficients

Thus, we have 12-3=9 degrees of freedom

Then we can use the 9 DF table of Student’s t to get a confidence interval

confint(model)
                 2.5 %   97.5 %
(Intercept)  4.1025560 6.162410
GenotypeKO  -0.6952039 1.683310
StressHigh   1.4400824 3.818597

Interpretation

                 2.5 %   97.5 %
(Intercept)  4.1025560 6.162410
GenotypeKO  -0.6952039 1.683310
StressHigh   1.4400824 3.818597

(Intercept) is the base expression level for Wild Type and Low Stress

GenotypeKO is the differential expression due to gene KO. It can be 0, so we do not have enough evidence to say “there is an effect”

StressHigh is the differential expression due to stress. It seems greater than 0 at 95% confidence. Will still be >0 at 99% confidence?

𝑝-values


Call:
lm(formula = logExpr ~ Genotype + Stress)

Residuals:
    Min      1Q  Median      3Q     Max 
-0.8916 -0.6917 -0.3380  0.8641  1.4262 

Coefficients:
            Estimate Std. Error t value Pr(>|t|)    
(Intercept)   5.1325     0.4553  11.273 1.31e-06 ***
GenotypeKO    0.4941     0.5257   0.940 0.371872    
StressHigh    2.6293     0.5257   5.001 0.000738 ***
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

Residual standard error: 0.9106 on 9 degrees of freedom
Multiple R-squared:  0.7421,    Adjusted R-squared:  0.6848 
F-statistic: 12.95 on 2 and 9 DF,  p-value: 0.002247

Expression v/s Fold Change

Volcano plot

Doing Multiple tests

The problem with multiple tests

Multiple test correction

There are several approaches

  • Family Wise Error Rate control (Bonferroni)
  • False Discovery Rate control (Hochberg)
  • Others

The basic difference is the trade-off between False Positives and False Negatives

Two ways of being wrong

In every hypothesis test, we can be wrong in two ways

  • Type 1: False positive
    • putting innocents in the jail
  • Type 2: False negatives
    • freeing criminals

Usually improving one means worsening of the other

How does FWER work

Family Wise Error Rate multiplies each p-value by the number of cases \[p.adj = p.value ⋅ N\]

It reduces False Positives and increases False Negatives

Sometimes we get nothing significant

How does FDR work

FDR sorts the p-values and multiplies each by an increasing value \[\text{p.adj} = \text{p-value} \cdot\frac{i}{N}\]

If we get \(p.adj<0.05\) then the probability that it is a false positive is 5%

Original v/s adjusted p-value

Looking for Biological meaning

Enrichment analysis

At this point we have a list of DE genes

Let’s say they are 10% of all genes

Each gene is part of one or more networks

  • Enzimes are in metabolic networks
  • Transcription factors are in regulatory networks
    • and regulated genes also
  • Other genes are part of signaling networks

We can draw expression on a network

Gene Ontology

Beyond pathways, it is useful to look at “Gene Ontology”

It has three branches

  • Molecular Function
  • Cellular Component
  • Biological Process

GO is a graph

Which networks shall we look at?

To find the relevant networks, we use Enrichment Analysis

For example, let’s say we have 10000 genes in total, and 1000 DE genes

Let’s say that carbon metabolism takes 3000 genes in total

If, among the DE genes, 300 genes are in carbon metabolism, we will not be surprised

Is this random or not?

3K carbon metabolism in 10K total genes is 30%

If we take 1K random genes, we expect to have 30% of them in carbon metabolism, just by randomness

If instead we have 600 carbon metabolism among the 1K DE genes, we are surprised

In that case we say that the set of carbon metabolism is enriched in the DE experiment

How to do enrichment analysis

This problems follows an hypergeometric distribution

\[ℙ(k \text{ observed sucesses} | N,K,n)\]

  • \(N\) is the population size,
  • \(K\) is the number of success states in the population
  • \(n\) is the number of draws (i.e. quantity drawn in each trial)
  • \(k\) is the number of observed successes

In our terms

\[ℙ(k \text{ DE genes in group} | N,K,n)\]

  • \(N\) is the total number of genes,
  • \(K\) is the total number of genes in the group
  • \(n\) is the number of DE genes
  • \(k\) is the number of DE genes in the group