What is the goal of 2D expression studies?

We want to find experimentally induced expression changes quickly, reliably and objectively.

The assumptions we can make to help achieve this goal are,

• Most expression studies are designed to vary as few spots as possible (usually less than 5%).
• Usually over 95% of spots are likely to not change due to experimental conditions.

So should we think backwards?

• Reliably and quickly reject the spots that aren't changing.
• Let the statistical tools do the work.
• Don't waste time fixing all the spots to find 'nothing'.

With the aim of finding experimentally induced expression changes, the answer is simply: Yes.

However, missing values present a big problem trying to achieve this with traditional analysis approaches, so we needed something different. We tried several other approaches, none of which were good enough. Then we created the SameSpots approach.

How does SameSpots achieve this?

During analysis, we look at all of the images together, find a single representative spot pattern and apply the same outline for each spot to the corresponding spot on every image in the experiment. This means we get a real measurement from each image, so we avoid the problem with missing values and we have 100% matching all of the time. There is no need to do any match editing.

Once we have the measurements from every image, we can use the statistics to trivially reject the spots that aren't changing. The few remaining spots that are changing can then be quickly validated.

How can we use the same outline for every image?

Highly accurate pixel-level image alignment allows us use the same spot outlines. In the same way that background subtraction and normalisation correct for measurement variation from gel to gel. Alignment corrects for positional variation from gel to gel. If we did not have accurate alignment at the image level, this approach would not work.

By aligning the images before performing spot detection, we can make sure that each pixel on an image corresponds to the same pixel at that location on every other image in the experiment. Alignment is the key part of the SameSpots workflow, and is the most important to get right. As well as spending a lot of effort on our alignment algorithms we also created specific alignment views to make validating alignment as easy and fast as possible

The screenshot below shows an overlay of the images being aligned. One in green and one in purple. Before alignment you can see a large difference in the location of the spots, but after alignment the spots overlay exactly (where green & purple combine to show black)