Algorithms {image processing algorithms} can determine background, cluster, check error and noise, find fiducials, normalize, check crosstalk, find signals, find features, aggregate replicates, and perform statistical analysis. Statistical cluster analysis can identify classes and assign shared feature. Hierarchical clustering can cluster into hierarchies. Self-organizing maps can group into equal categories.
Determining background {background determination algorithm} involves surface-fitting algorithm that detects global background changes across array. Background subtracts from features. Algorithm finds variance within feature integration aperture after global correction. Using average feature density can overestimate background. Putting space between feature blocks can find background.
Algorithms {error model algorithm} can check feature-signal noise, measure background variance, find cross-hybridization, check for spatial crosstalk and spectral crosstalk, measure normalization variation, and study replicates. Error model weights optimize signal-to-noise ratio. It has confidence value. Error flags label too-high variance, hybridization-control variance, high background, rejected pixels, bright neighbors, too-low signal-to-noise ratio, and saturated pixels.
Fiducials {fiducial-finding algorithm} are marks or shapes every 300 pixels, for automatic spot finding and row and column counting, without using quantitation-control scheme. Fiducials must be easy to distinguish from other image features and artifacts. Spotting and shrinking cause non-linear distortions and determine fiducial frequency needed. Row and column drift must be less than half the distance, five or six pixels, between features.
To account for labeling-amount, dye, fluorescent-detection, spotting, RNA-concentration, and sample-quantity differences, systems modify intensities {normalization, results}. Normalization allows comparison among slides and cell extracts.
types
Normalization can normalize on total intensity. Normalization can normalize on means and use ratio statistics. Normalization can use linear regression. Non-linear regression includes local regression, such as Locally Weighted Scatterplot Smoothing (LOWESS). Normalization algorithms use dilution-series controls, dye selection, filter selection, dye-quenching non-linearities, multiple gain settings, photobleaching amounts, and array-to-array normalization.
Strong signals at features can spread to neighboring features and skew weak signals. Features must be far enough apart to prevent more than 0.1% spatial crosstalk. Algorithms {spatial crosstalk mitigation algorithm} can reduce number of strong signals adjacent to weak signals. Algorithms can weight intensity sum across feature depending on neighboring features. Algorithms can use deconvolution with particular instrument.
Algorithms {signal integration algorithm} {signal quantitation algorithm} can use median pixel, integrate intensity, or use pixel-ratio median to generate values. First, algorithm aligns channels, if necessary. Fixed aperture is in both channels. Simultaneous scanning in both channels reduces crosstalk and allows pixel-to-pixel regression, which is more robust to defects. Integration can be only for high signal-to-noise pixel subsets.
Finding sites {spot-finding algorithm} can use three methods: find isolated peaks, align grid, or align centroid. To find isolated features or peaks, first shrink image to convolve it with feature model and then find isolated peaks. To align grid, first find fiducials and then fit rows and columns. To align centroid, first move off grid and find intensity-weighted centroid, if signal-to-noise feature ratio allows.
Algorithms {probe aggregation algorithm} {replicate aggregation algorithm} can average replicates.
3-Computer Science-Systems-Computer Vision-Algorithms
Outline of Knowledge Database Home Page
Description of Outline of Knowledge Database
Date Modified: 2022.0225