Example of use of XplorMed
This is the step that takes longer because it makes the computation of relations between words based on their correlated presence across the same abstracts (including the titles of the papers).
The next window displays a list of words ranked by a score that reflects their level of association to other words (see Figure 5a). The words analysed are only nouns. We make the analysis on the word lemma (e.g., "protein" and "proteins" are considered the same word).
|Figure 5a. Words ranked by their association score.|
|Figure 5a. Words ranked by their association score. You can explore the context and relations of the words listed by clicking on the corresponding links. There are two parameters (Alpha, and Score) that can be modified prior to doing the next step (see below). Notice the title at the top "iteration 1 - step 3". We are already in the iterative part of the process.|
Although at this point you have not been shown relations between words, this can give you already an idea of the main terms present in your abstracts. The next step will allow you to display some chains of related words selected from the stronger associations. You can control the way XplorMed is going to do that via two parameters whose values can be modified: Alpha and Score.
Basically, these parameters control the number of relations that are going to be displayed in the next step. The problem is that depending on the quality of your query, there are values for these parameters that may result in retrieving too many or too less relationships. That is why we cannot keep them fixed (and hidden), and that is why we have to offer to the user the possibility of playing with them.
All the relations from a word to another have an associated value of strength. Alpha is a threshold to be applied on the relationships to be displayed in the next step. If you raise it, you will get less connections. If you lower it, you will get more connections.
As you can see in the word list, each word has a score that indicates how much the word is associated to others. Score is a threshold to be applied in the words to be used for computing the relationships in the next step. If you raise it, then you will compute relationships on less words, and therefore you will obtain less relationships. Lower it, if you want to include more terms.
But, before proceeding to the next step, you can already explore the relationships of the words of the list. By clicking on a word you can see the sentences that contain that word in your selected abstracts (with links to those entries in MEDLINE). For example, you may see the context of the word app (at the middle of the list in Figure 5a) by clicking on it. In this way, you can already get an idea of why some abstracts are mentioning this. It turns to be out a beta amyloid precursor protein. (see Figure 5b).
|Figure 5b. Sentences containing the word "app".|
|Figure 5b. Sentences containing the word "app". This window displays only the sentences containing the word selected (in red).|
Abstract 9453569 indicates that 'heparin stimulates the synthesis and secretion of beta amyloid precursor protein app'. By abstracts 8863493 and 8158260 we learn that there exists heparin binding domains in app.
May be, more interesting information can be obtained by looking at the terms strongly related to a concrete term. By clicking at the [R] link at the right of a word in the list displayed in Figure 5a, you can explore the relation of the corresponding word to other words. For example, you could see the words related to app (see Figure 5c).
|Figure 5c. Words more strongly related to "app".|
|Figure 5c. Words more strongly related to "app". This window displays the words related to app. There are three levels of dependency. app is included in indicates that, for example, if "app" is used in an abstract, the words "protein", "alzheimer", or "precursor", will be also used. "app" depends on them, as for example a modifier depends on the word that is modified. app is always with indicates that two terms tend to occur always together in the same abstracts and never separated. This is rare. app includes indicates that the if the term "outgrowth" is used, then the term "app" will also be used. It depends on app.|
The new window with the related terms, includes also links that allow further dependency and context analysis. By clicking the words theirselves you can examine the sentences containing the word (as in Figure 5b), the [R] links indicate again that you can see the relations of the corresponding word to another words, and the [X] links give you the sentences containing both the corresponding word and the word that is related.
For example, you may not know what is the meaning and the (conceptual) relation of the terms that seem to depend on app (i.e. "outgrowth"). Then, just click on the [X] link to the right of the "outgrowth" word (as displayed in Figure 5c) and you will be shown the sentences of your abstracts containing either of the words, but from abstracts containing both words (see Figure 5d).
The value that is shown after the [R] and [X] links indicates the fraction of abstracts containing the dependent word where the other word was actually found together.
|Figure 5d. Context of the words app and outgrowth.|
|Figure 5d. Context of the words app and outgrowth. Sentences containing either of the terms (from abstracts containing both terms) are displayed in this window. The colouring can help to see the semantic relatedness of the terms. For example, in this case, there is a sentence (coloured in blue) that contains both terms, and even the terms appear contiguously (marked in pink). From this sentence on can easily realise that app is involved in the regulation of neurite outgrowth.|
Another more flexible way to obtain the same information is to use the window that pops out when you hit the "Explore context of any word" button at the bottom of the window displayed in Figure 5a) (see Figure 5e).
|Figure 5e. Word context window.|
|Figure 5e. Word context window. In this window you can type any word and find basically the same information as we did above with lipase and dependent words. The top window and button gives you the dependencies of the typed word, the middle one, the sentences, and the bottom one, the sentences from abstracts containing two terms. Note that if you type a word that was not analysed by the server (e.g., if you mistype, or if you type a verb) you will not obtain any result.|
Now, we will just do the next step (clicking the button "Compute chains of related words", Figure 5a) with the default values of Alpha and Score.