Splitting, Lumping and Clustering

David circa 2009
David happily watching his videos in the summer of 2009

If we want to find treatments for Phelan McDermid syndrome (PMS), first we need to figure out what is PMS.  That was spelled out in my blog: Why don’t we have better drugs for 22q13 deletion syndrome? My next blog addressed how to organize all the different genetic deletions and mutations so that we can define PMS (Defining Phelan McDermid syndrome).  Today’s blog addresses ways we can define different types of PMS.  If we don’t define different types, we are wasting our time experimenting with treatments.  For instance, some PMS kids talk fluently, some talk in short sentences, some can only say single words and many, like David, do not talk at all.  These and many other difference warrant different groups of kids when we test treatments.

Just as there is huge variation in abilities and behavioral characteristics, our kids have very diverse genetics.  Recent studies of rodents  show that not all with Shank3 mutations are alike.  In fact, drugs may work very differently on different Shank3 mutations. Anyone who has kept up with my blogs knows that deletions of different genes are likely to have very different effects on our children.  These difference are very important.

Useful drug testing is stuck right now until we develop a way to categorize people with PMS based on both phenotypic characteristics (symptoms and manifestations) and genotypes (deletions versus mutations and which genes are affected).

I have heard scientists who study Shank3 mice talk about “splitting” and “lumping”.  Splitting is breaking groups into subgroups.  Lumping is putting everyone/everything together into a single group.  Lumping has not worked and the growing consensus is that lumping will never work in our population. Splitting based on just one characteristic (e.g., deletion size) probably won’t work, either.  We need a more refined approach.  What we need is “clustering”.  Clustering is what mathematicians and scientists do when categorizing requires using many different characteristics at once.

Here is an example.  Let’s say you want to buy a car.  You might look at various cars and think about both price and gas milage.  You could make a graph something like this:

Car Clustering2Similar types of cars have similar prices versus gas milage tradeoffs.  Race cars are more expensive, but get poor gas milage. Clustering is when you identify meaningful subgroups on a graph because the individual points are close together.  Each group is a cluster.  Even if not every car fits neatly into a cluster, you still have an organizational scheme that can be very helpful.

PMS needs meaningful groups.  Clustering can get complicated when there are more and more features that divide up the population.  However, computer programs can take care of the complexities.  What we need first is to identify which characteristics are important for grouping.  As a practical matter, researchers go back and forth. They consider characteristics, run a program that automatically clusters the data based on those characteristics, and then look to see if the clusters make sense.  That is what we need to do.

When we took David to the PMS Foundation Family Conferences in 2008 and 2010, we met a handful of kids that were remarkably like David (see photo of David, above).  What was it about those kids?  As I recall, they walked the same way, loved watching music videos, asked for help the same way, were nonverbal and all have relatively larger deletions.  Are those meaningful characteristics?  Will they help us divide PMS into different groups for meaningful drug studies? We need to find out.

arm22q13

 

Previous blogs

Defining Phelan McDermid syndrome
Why don’t we have better drugs for 22q13 deletion syndrome?
What do parents want to know?
Is 22q13 deletion syndrome a mitochondrial disorder?
Educating children with 22q13 deletion syndrome
How to fix SHANK3
Have you ever met a child like mine?
How do I know which genes are missing?

Mouse models
Science Leadership
How can the same deletion have such different consequences?
22q13 and the hope of precision medicine
22q13 Deletion Syndrome: hypotonia
Understanding gene size
Gene deletions versus mutations: sometimes missing a gene is better.
Is 22q13 deletion syndrome a ciliopathy?
Understanding translocations in 22q13 deletion syndrome: genetics and evolution
Understanding deletion size
Can 22q13 deletion syndrome cause ulcerative colitis?
Can 22q13 deletion syndrome cause cancer?
22q13 deletion syndrome – an introduction

Advertisements

3 thoughts on “Splitting, Lumping and Clustering

  1. I think we should start off by splitting the verbal and nonverbal kiddos . You posted how much David loves the Happy Birthday song and thats Jenna’s favorite also. We sing it to get her to smile for pictures. She either likes the tune or she equates it to eating cake !

    Like

  2. Your blog makes perfect sense to me. I’m not a geneticist but the differing amounts of deleted material explains the enormous differances in each persons characteristics and abilities. Thank you for your dedication to our group. It is very much appreciated.

    Like

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s