Medical Subject Headings (MeSH) analysis

class: center, middle, inverse, title-slide

# Medical Subject Headings (MeSH) analysis
## LIFE 891-002: Integrating Quantitative and Computational Biology into Life Sciences Research
### Gota Morota <br /><br /> <a href="http://morotalab.org/" class="uri">http://morotalab.org/</a>
### 2018/02/28

---

# Medical Subject Headings (MeSH)

- collection of a comprehensive life sciences vocabulary

- used for indexing articles in MEDLINE database (PubMed)

- includes over 26,000 terms (as of 2011)

- 10 `\(\sim\)` 15 MeSH headings per article

<div align="center">
<img src="meshlogo.png" width=800 height=300>
   <figcaption>NCBI MeSH website</figcaption>
</div>

---

# Medical Subject Headings (MeSH)
<div align="center">
<img src="canavez2012PubMed.png" width=800 height=450>
   <figcaption>Paper in PubMed</figcaption>
</div>

---

# Medical Subject Headings (MeSH)
<div align="center">
<img src="canavez2012PubMed2.png" width=500 height=450>
   <figcaption>MeSH terms</figcaption>
</div>

---

# MeSH term-based search
Enhances PubMed search
<div align="center">
<img src="pubmedMesh1.png" width=800 height=450>
   <figcaption>MeSH terms</figcaption>
</div>

---

# MeSH term-based search (cont.)
Enhances PubMed search
<div align="center">
<img src="pubmedMesh2.png" width=800 height=450>
   <figcaption>MeSH terms</figcaption>
</div>

---

# MeSH Categories
<div align="center">
<img src="table1.png" width=800 height=450>
   <figcaption>Tsuyuzaki et al. (2015) </figcaption>
</div>

---
# Gene Ontology (GO)
GO terms are assigned to genes according to

- biological process

- molecular function

- cellular component

It helps to understand biological interpretation of genes, however,

- GO is restricted to specific ontology

- MeSH contains terms regarding other concepts such as Diseases, Chemicals and Drugs, and Biological Phenomena

---
# MeSH enrichment analysis
**Hypergeometric test**
<div align="center">
<img src="hypergeometricEq.png" width=400 height=150>
</div>

- `\(S\)` is the total number of selected genes

- `\(N\)` is the total number of analyzed genes

- `\(k\)` is the total number of genes in the MeSH term under study

- `\(g\)` is the total number of selected genes in the MeSH term

---
# Hypergeometric test 
<div align="center">
<img src="MeSHhyper.png" width=900 height=650>
</div>

---

# Hypergeometric test (Example)
The p-value is the probability of getting `g` or more annotated selected genes in a sample of size `S` from a sample of background genes with `k` annotated genes and `N-k` non-annotated genes.

```r
S <- 300
N <- 20000
k <- 2500
g <- 50
# 1
sum(dhyper(g:min(S,k), k, N-k, S))
[1] 0.02031964
phyper(g-1, k, N-k, S, lower.tail=FALSE)
[1] 0.02031964
# 2
1 - sum(dhyper(0:g-1, k, N-k, S))
[1] 0.02031964
1 - phyper(g-1, k, N-k, S)
[1] 0.02031964
```

---
# Number of background genes matters

```r
S <- 300
k <- 2500
g <- 50

N <- 10000; 1 - phyper(g-1, k, N-k, S)
```

```
## [1] 0.9998374
```

```r
N <- 15000; 1 - phyper(g-1, k, N-k, S)
```

```
## [1] 0.5245452
```

```r
N <- 20000; 1 - phyper(g-1, k, N-k, S)
```

```
## [1] 0.02031964
```

---
# MeSH ORA framework in Bioconductor
<div align="center">
<img src="Fig1.png" width=800 height=450>
   <figcaption>Tsuyuzaki et al. (2015) </figcaption>
</div>

---
# Entrez Gene ID `\(\longleftrightarrow\)` MeSH ID
<div align="center">
<img src="Fig6.png" width=800 height=450>
   <figcaption>Tsuyuzaki et al. (2015) </figcaption>
</div>

---
# Bioconductor packages
Cattle annotation package 
<div align="center">
<img src="meshBtaegdb.png" width=800 height=200>
</div>
Statistical analysis package
<div align="center">
<img src="meshr.png" width=800 height=210>
</div>

---
# Example code

```r
library("meshr")
library("MeSH.db")
library("MeSH.Bta.eg.db")

meshParams <- new("MeSHHyperGParams", geneIds = my.geneID3, 
                 universeGeneIds = univ.geneID3, 
                annotation = "MeSH.Bta.eg.db", 
                category = "D", database = "gene2pubmed", 
                pvalueCutoff = 0.05, pAdjust = "none")

meshR <- meshHyperGTest(meshParams)

head(summary(meshR))
MESHID Pvalue OddsRatio ExpCount Count Size    MESHTERM            GENEID SOURCEID
D015815 5.00e-06  12.2754 0.696    7    39 Cell Adhesion Molecules 404118 23800882
D015815 5.00e-06  12.2754 0.696    7    39 Cell Adhesion Molecules 538486 15117967
D015815 5.00e-06  12.2754 0.696    7    39 Cell Adhesion Molecules 281485 15117967
```

---
# Visualization for MeSH enrichment analysis

```r
tagcloud(mesh.tags, weights=weights, col=colors, order="size") 
```
<div align="center">
<img src="meshtags.png" width=700 height=500>
</div>

---
# MeSH analysis in mammalian livestock
Morota et al. (2015)
<div align="center">
<img src="Morota2015.png" width=800 height=230>
</div>

- Dairy Cattle - genes that showed differential expression in preimplantation embryos due to maternal methionine supplementation (Penagaricano et al., 2013)

- Swine - genes related to age at puberty in swine (Tart et al., 2013)

- Horse - genes under selection in Quarter Horse (Petersen et al., 2013)

---
# Number of annotated genes in cattle, swine, and horse
<div align="center">
<img src="Table1mesh.png" width=800 height=400>
   <figcaption>Morota et al. (2015) </figcaption>
</div>

---
# Dairy Cattle - RNA-Seq
<div align="center">
<img src="MeSHdairy2.png" width=900 height=490>
   <figcaption>Morota et al. (2015) </figcaption>
</div>

---
# Swine - GWAS
<div align="center">
<img src="MeSHswine.png" width=800 height=490>
   <figcaption>Morota et al. (2015) </figcaption>
</div>

---
# Horse - Selection signature
<div align="center">
<img src="MeSHhorse.png" width=800 height=490>
   <figcaption>Morota et al. (2015) </figcaption>
</div>

---
# MeSH analysis in chicken
Morota et al. (2016)
<div align="center">
<img src="Morota2016.png" width=800 height=310>
</div>

- Differential expression in abdominal fat tissue between high and low feed efficiency broiler chickens (Zhuo et al., 2015)

- Candidate genes historically impacted by selection in 72 different chicken breeds (Beissinger et al., 2015)

---
# MeSH annotation of the chicken genome
Number of known and selected genes annotated by MeSH (Medical Subject Headings) and GO (Gene Ontology).

---
# Significant MeSH terms

---
# MeSH semantic similarity - information content

- semantic similarity among MeSH terms

- semantic similarity among genes

- information content-based measure

<div align="center">
<img src="meshHierarchy2.png" width=800 height=390>
   <figcaption>Morota et al. (2016) </figcaption>
</div>

---
# MeSH semantic similarity (RNA-seq)
Chemicals and Drugs
<div align="center">
<img src="rnaseqMeSHSimD.png" width=800 height=490>
   <figcaption>Morota et al. (2016) </figcaption>
</div>

---
# MeSH semantic similarity (selective sweep)
Chemicals and Drugs
<div align="center">
<img src="epiMeSHSimD.png" width=800 height=450>
   <figcaption>Morota et al. (2016) </figcaption>
</div>

---
# Gene semantic similarity (RNA-seq) 
<div align="center">
<img src="rnaseqGeneSimMeSHBPRevised1.png" width=850 height=400>
   <figcaption></figcaption>
</div>

---
# Gene semantic similarity (selective sweep)
<div align="center">
<img src="epiGeneSimMeSHBPRevised1.png" width=850 height=400>
</div>

---
# MeSH analysis in maize
Beissinger and Morota. (2016)
<div align="center">
<img src="Beissinger.png" width=800 height=310>
</div>

- regions under selection during maize **domestication** (Hufford et al., 2012)
- regions under selection during maize **improvement** (Hufford et al., 2012)
- regions under selection for **seed size** (Hirsch et al., 2014)
- regions under selection for **ear number** (Beissinger et  al., 2014) 
- regions contributing to **inflorescence** traits (Brown et al.,2011)

---
# MeSH enrichment analysis (maize)
<div align="center">
<img src="MeSHmaize.png" width=700 height=300>
   <figcaption></figcaption>
</div>

---
# MeSH semantic similarity (maize)
<div align="center">
<img src="SemanticSimilarity.png" width=800 height=500>
   <figcaption></figcaption>
</div>

---
# Pitfalls

What are the pitfalls in MeSH / GO analysis?

- significant MeSH or GO terms can arise from a random set of genes

- easy to make biological sense out of the false-positives

- inferring the molecular mechanism on the basis of statistically significant MeSH or GO annotations alone is  potentially error-prone

Literature

- Pavlidis et al. (2012). A critical assessment of storytelling: gene ontology categories and the importance of validating genomic scans. MBE. [doi:10.1093/molbev/mss136](https://doi.org/10.1093/molbev/mss136)

---
# Summary 
MeSH

- share similar concepts derived from GO annotations

- able to draw additional information not directly provided by GO

- supplementary tool to GO analysis

- <span style="color:red">good tool for generating hypotheses</span>