GeneNetwork: A tour and tutorial

Aim of this tutorial. The goal is to illustrate how to use GeneNetwork (GN) to study gene function and relations between genes and traits such as differences in disease severity, differences in anatomy, and differences in physiology and behavior. You can also use GN to study relations among phenotypes, for example: to what degree does an injection with cocaine or alcohol lead to an increase in movements? Most of the experimental data sets in GN are from small populations of mice (e.g., the BXD family of strains) and rats (HXB family). There are also some human, monkey, drosophila, and plant data sets to explore, although you may find that mapping functions have not yet been implemented for these other species.

The focus in this tutorial is how to use some of the most important functions. You should be able to complete this tutorial in about an hour. For this demonstration we will study expression of a key gene known as NR2B or Grin2b. This gene/mRNA/protein is crucial in learning and memory. Once you have worked through this example, you should be able to use GN to explore single genes or set of genes, mRNAs, and other standard traits that interest you.

What you will learn. If you spend an hour working through this tutorial you will learn how to extract dozen of molecules that potentially interact with NR2B. It will be easy for you to generalize what you learn to any other gene or transcript of interest. You should be able to confirm known relations (information from the literature) and you should be able to uncover intriguing new relations among sets of molecules and other traits. You will learn how to exploit a gene ontology (a gene ontology is a simple and systematic way of categorizing the functions of genes) and you will also learn about gene or QTL mapping and complex trait analysis.

By the way: if a term is new to you (What is a QTL?), and you would like to read an explanation then have a look at the GeneNetwork Glossary.

Much of the data in GN were gotten from gene array experiments. As you have undoubtedly heard or experienced yourself, the analysis of array data sets is difficult and sometimes messy. You may encounter poor quality data in some of these enormous data sets. An important part of learning how to use GN involves tools to evaluate data quality. Treat the results you generate with GN with caution. There are solutions to some problems you run into, but for other problems, including comparison across multiple data sets.

After you have worked through with this tour, please look quickly at the Frequently Asked Questions and the Glossary. Let me know if you have other questions or if you see mistakes that I should fix. Email should go to Rob Williams (rwilliams@uthsc.edu).

Step 1: Getting the terminology right. The use of gene symbols and names in research papers is not consistent. When you search databases it helps to use the preferred or official gene name and symbol. In most papers, the NMDA 2B receptor is abbreviated NR2B and a Google search for "NR2B" will generate about 100,000 hits, many of which deal with the genetics and molecular biology of learning and memory. But it turns out that NR2B is not the official name of either the gene or protein. A great way to verify nomenclature is to go to NCBI Entrez Gene and enter the gene name or symbol. Entrez gene should be able to resolve your query and give you the correct symbol.

Figure 1: Getting the terminology right

When we enter NR2B or nr2b we find that the official gene symbol in Entrez Gene is Grin2b for mice. The corresponding human gene is the same, but written entirely in capital letters GRIN2B. In both species the official gene name is glutamate receptor, ionotropic, N-methyl D-aspartate 2B.

Step 2. Linking to the GeneNetwork search page Link to GN at www.genenetwork.org/. Ideally, keep this tutorial page open at the same time so that you can look back and forth between the two windows. You have a few choices to make: Choose species = Mouse, Group = BXD, Type = Hippocampus mRNA. We have a great deal of data for the BXDs, so when in doubt, please select the BXD group of mice. Having made these choices, you still need to pick a particular database; in this case an array data set for a particular brain region called the hippocampus that, like NR2B, is critical in learning and memory. For the purpose of this tutorial, choose the database file called:

Hippocampus Consortium M430v2 BXD (Jun06) PDNN

You can set your particular choice of species, group, type, and database as your personal default setting. Simply click on the Set to Default button (lower right). If you want to know what this long database term is all about, click on the INFO button immediately to the right of the database name.

Now enter your search terms in either of the search term fields labeled Get Any or Combined. Get Any is usually best and will search for all of the entries you put in this field (logical OR). Combined will only get records that match all of the terms that you enter (logical AND). You could enter both NR2B and Grin2b in the ANY field. You can also use wildcard characters ? and * for single or multiple characters. It is often a good idea to enter an asterisk after a search term, such as Grin*. This will get all subunits of a molecule or complex.

Figure 2: The Search Screen. This is the page to bookmark.

Step 3. Retrieving the data. When you click on the Search button, your computer sends this string of search terms to GN (GN is an Apache-Python-MySQL web database system), which then looks through thousands of records for matching terms. The database that we just searched has 45,101 entries that represent close to 20,000 known genes and expressed sequence tags (ESTs). (To determine the size of any database enter a single asterisk (*) in the ANY field.)

Figure 3: Search results for the search string Grin2b

In this particular case, if you entered just "Grin2b", your will get at a list of six data sets. The last three are measurements of Grin2b expression. These thee measure different parts of the mRNA: the distal 3 prime UnTranslated Region (3' UTR), the 3' region of the last coding exons (coding exon 12), and an alternative 3' UTR of a short mRNA splice variant (an mRNA isoform). Which of the three should you pick? The best choice is usually the that which corresponds to coding sequence. In this case, you should study the 5th entry highlighted in red. You an come back to the other two data sets later to see how they compare. And you can ignore the top three Kif17 data sets that were found only because the gene description includes the text "NR2B/GRIN2B NMDA receptor transporter."

This is a good point to review several of the features most GN pages. Feature 1 is the banner of terms toward the top labeled Home, Search, Help, News, etc. etc. Most of these menu headings have pop-down lists from which you can select additional resources and tools. For example, the Search menu heading lists the Search Databases page (our starting point), the SNP Browser tool, the GeneWiki resource, Interval Analyst, and GenomeGraph displays. These features are worth trying out later. The Policy menu explains how to contact the data providers and how to use and cite data.

Another useful feature of the Search Results window is the Sort By selector. You can sort longer lists of "hits" by their location, their expression levels, or by their maximum LRS or LOD scores. There are also small check boxes to the left of each entry. These are used to select data you would like to move into a Collection. The Add button will move any checked items into your collection. Collections can include any gene, trait, or SNP marker that has measured in the BXD family. You can even add your own BXD data using Enter Trait Data in the MAIN menu (top left). The ability to add diverse data types into a Collection provides a great deal of power. The use of the Collections is a topic for a more advanced tour. Let's get through this quick tour and then feel free to build up your own collections of phenotypes for "collective" analysis.

Step 4. Reviewing the NR2B expression data. Roll you cursor over the term ProbeSet/1422223_at in the Search Results window. This is the probe set that targets the last exon of the Grin2b transcript. The text will turn red. Click on the term. This will generate a new page called the Trait Data and Analysis Form. This Trait Data and Analysis Form page is the most important page for the analysis of genes and traits. We will return here several times. The top of this page contains useful background information, including the database that we used, the trait identifier, gene symbol and aliases, the chromosomal location and megabase position (Mb) of Grin2b in the mouse genome. GN also includes links to NCBI, OMIM, GenBank, BioGPS, STRING, PANTHER, Gemma, and the Allen Brain Atlas (ABA). To find out more about these resource, just click on the links. There are also many additional useful links under the Links menu heading.

Figure 4: Trait Data and Analysis Form for Grin2b, probe set 1422223_at

Eight buttons shown in Figure 4 and you need to know what most of them do.

1. SNP Variant Browser: This link provides you with a list of all known SNPs in Grin2b that are in the GN database. You can, of course, also search for SNPs in other genes or regions. The SNP Variant database is pretty well populated (about 8 million SNPs) and includes all of the Celera SNPs and many SNPs from Perlegen, NIEHS, and our own in-house sequencing projects. As of Jan 2011, there are nearly 7000 SNPs in Grin2b, but only about 20 of these are in exons. Click on the SNP Browser button to have a quick look 9you may need to restrict the SNP search to just Domain = Exon.
2. GeneWiki: This link lets you annotated our databases. You can leave yourself notes and comments about particular genes or probe sets. You can easily find your own notes using a special search string described in the Advanced Search page. But in short your search would be written out "wiki=myName". Leave out the quotes and make sure that "myName" is in your Wiki entry. It is that simple.
3. Verify Location and Verify RNA Seq: These buttons are used to confirm that the data set correctly targets the last exon of Grin2b. Click on either link. The probe sequences used on the array are sent to the Genome Browser and the best match is found in real time using the BLAT algorithm. The Search Results page allows you to drill down to a view of the genome. Click on the "browser" link to the far left (top row). Look for the horizontal track that is made up of a series of black rectangles labeled "Blat Sequence" or "Probe XXXYYY". You will also see several "tracks" labeled Grin2b and GRIN2B. If you use the Zoom Out 10X button you will see that the probes and probe set are aligned with the 3' end of the last exon--a bit more detail than we had before. You may also notice that Grin2b is encoded on the minus strand of chromosome 6 and that the tiny arrow heads visible on the last intron point to the left (the transcription direction is from right to left). The Verify RNA seq button has the same function, but takes you to a GN mirror of the UCSC Browser that has added RNA sequencing data for brain, hippocampus, and eye (Jan 2011) of many BXD strains of mice.
4. Basic Statistics: This button will generate summaries such as the average expression, the range, bar charts of expression ordered by strain and by rank. Try it quickly.
5. Similar Traits: This button will provide you a link to Grin2b expression data in other data sets that may interest you.
6. Probe Tool: A link to the sequence data for the individual probes that make up a probe set. This table can be used for a very fine-grained analysis of particular probe sets.
7. Add to Collection: If you would like to add a trait to your collection of traits, transcripts, or markers, use this button. This is the same function we mentioned earlier.
8. Reset: GN allows you to modify values for traits and this button is used to reset to the original values.

In general, text that uses a blue font is also a link. For example, the text at the top of Figure 4 Hippocampus Consortium M430v2 BXD (Jun06) PDNN will link you to a Materials and Methods "metadata" or Info page. There is lots of information on the Info page, but the short summary is that the NR2B expression data in the Hippocampus mRNA database we are exploring was generated from approximately 1200 hippocampii and 600 mice belonging to 99 strains (typically three animals per array and approximately 200 arrays). This is one of the largest data sets in GN. Each array includes samples from a single age, sex, and litter. The BXD strain family were all made by crossing two parental strains, C57BL/6J and DBA/2J. Both of these parental strains have been fully sequenced. The Hippocampus data set includes expression estimates for both parental strain, and also data for 15 other common inbred strains, for example, 129S1/SvImJ, C3H/HeJ, CAST/EiJ, and others. There is also a complementary, but smaller Hippocampus Consortium data set for the CXB strains of mice.

Farther down the page we encounter sections labeled Trait Correlations and Interval Mapping and Trait Data. We will come back to these tools in a moment, but keep scrolling down to the actual numerical data on gene expression for Grin2b. The larger numbers in the boxes (6.631, 6.612, etc.) are estimates of the abundance of Grin2b mRNA in the hippocampii of different samples of mice. The smaller numbers (0.184, 0.205, etc.) are the standard errors of expression, usually based on two arrays (hence SEM also equals SD). Numbers are all expressed using a log base 2 scale. A value of 8 therefore corresponds to 2^8 or 256. A difference of one unit is roughly equivalent to a two-fold difference in expression in Grin2b expression. DBA/2J has an expression of 6.506 +/- 0.008 whereas BXD14 has an expression of 7.604 +/- 0.232. That amounts to approximately a 2-fold difference in the amount of transcript.

These kinds of preliminary results often generate intriguing and testable hypothesis: do strains of mice have the anticipated differences in learning and memory performance given the known effects of Grin2b overexpression in transgenic mice? If you scroll down farther in this lists of Grin2b expression estimates you will see that the strain of mouse called BXD80 only expresses 6.098 units of NR2B whereas BXD42 expresses 8.400 units. That amouts to a putative 5-fold difference in mRNA level. We need to take this with a grain of salt, because these expression estimates are lower than we expect. The average expression for all transcripts (including those NOT expressed) is 8 units. The fact that the average expression of the Grin2b probe set is less than average should make us worry. Are these data too noisy to use? Is there an unsuspected problem with the data handling or the probes? These are hard questions to answer but the Probe Tool is useful because you can look at the expression of the individual probes values used to generate the probe set summary value. We will skip this process for now, but remember that this "deeper" level is always just one click away. For the time being, let's evaluate the data using the Basic Statistics function.

Figure 5: Basic statistics for Grin2b

Step 5. Basic Statistics.  To get a better understanding of these values and how expression estimates are distributed click on the Basic Statistics button. A new window will open with a statistical summary table and a box plot toward the top (Figure 5), two bar charts in the middle, and a normal probability plot toward the bottom. You may not be familiar with these types of plots yet, but they are simple to read. The box plot is a simply summary of the spread of the 86 values. The blue plus sign represents the mean expression. The box defines the 25% and 75% quantiles (if we had studied exactly 100 strains these would be those strains at the 25 and 75 rank. If you want more information, just click on the link beneath the plot. The bar charts are easy to read. They provide a graphic output of the data that you saw in the Trait Data and Analysis Form. The Y axis of the graph is truncated and does not extend down to a value of zero. This tend to highlight the variation within strains. The error bars are quite large, and are only based on two samples (in this case the SEM is usually the same as the SD). Also note that the size of these error bars tend to increase as the expression increases (non-uniform error). High noise and non-uniform variance are all characteristics that should reduce your enthusiasm. But let's persevere, because these data have not yet cost you a penny and because this is a great lesson.

Figure 6: Normal Probability plot for Grin2b

Toward the bottom of the Basic Statistics page you will find a Normal Probability plot (Figure 6). There is another link associated with this plot that will provide more background, but here is a brief explanation of how to read these plots. On the X axis is the expected Z score for every strain based on its ranking out of 86 strains. Values range from about -2.5 to +2.5. If you randomly drew 86 values from a normal distribution you would expect the lowest value to be about -2.5 standard deviations below the mean (or -2.5 Z) from the mean. The Y axis provides a read-out of the actual expression level. If the expression of NR2B were normally distributed then the strain averages would form a straight line. If expression of Grin2b were obviously controlled by a single Mendelian factor, then this plot would have an S shape with many high strains and many low strains and few strains with intermediate values. Instead this plot highlights skew toward low values (also seen in the box plot). A few strains have comparatively high expression, but the main feature is the excess of strains with values from 6.0 to 6.5 units. It does not look like a Mendelian trait. But looks can deceive.

What this plot also highlights is the wide range of expression of NR2B gene transcript in normal strains (the BXD strains are not mutant or knockout mice). But the high error raises the possibility that much of this variation is simply sampling error. If we performed an analysis of variance with Strain as our main effect, this data set would be associated with a modest and statistically insignificant F values. But we have other methods to evaluate this putative strain difference. We can map it and see if any interesting patterns emerge from the mapping that might cheer us up and demonstrate that the variation of strain means is actually true signal. In this case, the answer is (fortunately) a strong Yes (LOD = 16). But when you see data of this type, the usual answer will be No.

Variation in NR2B gene expression is a signal that we can now cautiously use to search for transcripts that co-vary. Does NR2B message covary with other subunits of the NMDA receptor complex (over 1000 transcripts are part of the postsynpatic density of which NR2B is a key member). Does the the expression of other genes compensate for the apparent 5-fold range in NR2B message level?

Step 6. Covariation of expression. To answer these types of questions return to the Trait Data and Analysis Form that has all of the values for Grin2b in 86 different strains of mice. This time select the Trait Correlation button. There are five pop-down menus that allow you to modify search parameters. Let's modify one of the default settings. Change Return to read top 500. The other seetings are fine for the time being.

Figure 7: How to set up the search for covariates of Grin2b

Within a few seconds of clicking this button, GN will return a new page of data, a Correlation Table of the top 499 transcripts that covary with variation in Grin2b expression. At the top of this list (sorted by p value) is Grin2b itself. The third best covariate of our Grin2b probe set is another Grin2b probe set. That is reassuring.

Let's review the columns:

1. The first column is just an index with check boxes. You can easily add items into your BXD Collection using these checkboxes.
2. Record ID: The ID of the trait; in this case just the probe set identifier given by Affymetrix
3. Symbol: The official gene symbol. Clicking on these symbols will link you to NCBI.
4. Description: The name of the trait or the name of the gene from which the mRNA is transcribed
5. Chr: The chromosome of the gene from which the transcript is transcribed
6. Megabase: The chromosomal nucleotide position of the most proximal end of the probe set (mm6 alignment)
7. Mean Expression: The average expression of the probe set (mean of strain averages)
8. Correlation: The correlation of with the reference trait, in this case with Grin2b probe set 1422223_at.
9. N Cases: The number of strains involved in the correlation analysis
10. p Value: The p value associated with the correlation and number of cases without correction for multiple tests.
11. Lit Corr: The Literature Correlation. This is a very cool column of data generated by Ramin Homayouni, Michael Berry and colleagues that summarizes the correlation of Grin2b with many other genes based upon an analysis of the PubMed Literature.