Limiting 'noise' in 2D gels

2D gel electrophoresis (2DE) is a scientific technique that is a cornerstone of proteomics research. 2DE has evolved dramatically both in terms of the number of scientists utilising the technique and how it is applied within their research. As a result, new technology is continually being developed and implemented to meet the growing needs of proteomics researchers. A key objective for researchers involved in high throughout proteomics is to remove subjectivity from 2DE analysis while processing large numbers of gels per experiment. To achieve this requires controlling variable gel running and gel image quality, providing high sensitivity for low level protein expression and being able to identify real changes in protein expression from large amounts of data. This enables researchers to focus on characterising novel drug targets.

Technology designed to meet these challenges has been developed within Progenesis Discovery. Its features provide advanced algorithms with high levels of automation and include: noise filters (INCA), combined detection, warping and matching steps, data quality control and statistical measurements.

INCA (Intelligent Noise Correction Algorithm) is a feature of Progenesis Discovery from Nonlinear Dynamics. By assessing noise at each pixel in the gel and accurately removing noise from all areas of the gel, including that within spot material, INCA significantly improves spot detection.

In Progenesis Discovery, the filtering stage can either occur separately, before image analysis, or as part of the automatic analysis during spot detection. Spot outlines are derived from the filtered images to drive highly accurate and robust spot detection and subsequent steps in the analysis process. The spot volumes can be measured from either the raw tif image, or from the corrected image (INCA volume). This means users can see for themselves the percentage of any spot volume attributed to noise and choose whether to view data with or without INCA processing applied.

All gels will contain some degree of 'noise'. For our purposes we can define 'noise' as everything that is not signal ie, any feature that does not relate directly to bound protein. The nature and difficulty of running 2DGE means that some degree of noise will be unavoidable.

The bulk of the noise in 2D gels can be broadly categorised into two main streams; low level noise ie, very small ripples and bumps across the whole surface of the gel and noise spikes ie, high amplitude points usually caused by dust and crystallised stain artefacts.

Other approaches used to remove noise make assumptions about the noise properties and process the whole of the gel.

For example, to remove the noise spikes it is usually assumed that they are high frequency and so some form of low pass filtering is applied (FFT, median mask, averaging mask, blurring filter) across the whole of the gel.

This means that even though you may reduce the noise (these techniques are unlikely to totally eliminate it) you can also distort the spots themselves.

A low pass filter does not differentiate between spikes and spots with steep sides! It is also difficult to get these techniques to automatically adjust to a given gel. Applying a median filter to a clean gel can still distort the spots.

Overcoming the problems

INCA performs a detailed multi-stage noise assessment at each pixel in the gel. It first identifies the noise from analysing the gel itself. It then adaptively applies specialised local corrective procedures that operate on the noise alone. This means that INCA does not do anything to a 'clean' gel. INCA has been designed to reduce noise while minimally affecting the spots themselves. You can verify this for yourself by selecting a small section of a noisy gel within Progenesis Discovery and viewing it in 3D. Within the software itself you can rapidly toggle between the original gel and the INCA version.

Surely my original scan is the only valid scientific record

Sound statistical experimentation has two phases: exploratory data analysis (EDA) and hypothesis testing (HT). EDA is used to formulate a hypothesis that allows a statistically robust experiment to be designed and tested. EDA experiments are usually characterised by small numbers of gels with very low numbers of repeats and/or using non-optimal running conditions. These features mean it can be unreasonable to expect a statistically significant result from an EDA experiment. The results of the EDA experiment are used to form a hypothesis (eg, compound X modulates the expression of protein Y) from which the scientist can design a statistically sound experiment to test the hypothesis.

INCA is designed to help find results in exploratory data. It can remove noise from samples as small as a single gel and maximise the chances of finding interesting results to test with further experimentation. A second use of INCA is in the analysis of 'one shot' gels where producing enough gels for statistical robustness is not an option (eg, brain biopsies). INCA can be invaluable in these situations as it may give the scientist all of the information needed from incredibly noisy data that cannot be repeated. Again, it's important to stress that the raw image is not afffected so the original scan is what users view with the option to record results with or without INCA processing applied.