= RNA integrity number
{c}
The \b[RNA integrity number] is an algorithmic score of <RNA integrity> inferred from an <electropherogram>. Its original scale runs from 1 for strongly degraded preparations to 10 for high integrity. The method developed by <Andreas Schroeder> and colleagues combines several trace features using a trained <neural network> <regression model>, rather than using a single ribosomal-peak ratio.<The RIN: an RNA integrity number for assigning integrity values to RNA measurements by Andreas Schroeder and others>[Abstract, Feature selection and Model training]{ref}
The original apparatus separates RNA in microfluidic channels and detects it through <fluorescence>. The resulting <electropherogram> reports signal against migration time. Peak and region features summarize the size distribution; the trained model maps that feature vector to an integrity score.<The RIN: an RNA integrity number for assigning integrity values to RNA measurements by Andreas Schroeder and others>[Techniques for measuring RNA integrity and Figure1]{ref}
An original area calculation illustrates why a shape feature differs from total signal. If a selected region has integrated signal $A_r=60$ and the whole trace has $A=100$ in consistent units, its area fraction is $A_r/A=0.60$. Multiplying every signal value by two gives $120/200=0.60$. Adding signal in a different fragment-size region can instead change the fraction. This simple ratio is not the RNA integrity number algorithm; it illustrates how a trace can contain information distinct from total material.
Training a score also requires checking predictions on measurements not used for fitting. Table1 of the original study partitions 1339 electropherograms into 937 training and 402 test records.<The RIN: an RNA integrity number for assigning integrity values to RNA measurements by Andreas Schroeder and others>[Table1: electropherogram counts]{ref} For a constructed error example, predicted scores $7.5$ and $8.5$ against labels $8$ and $8$ have mean squared error $(({-0.5})^2+0.5^2)/2=0.25$. That evaluates agreement with labels; it does not count intact molecules.
An RNA integrity number of 8 therefore cannot be converted into an $80\%$ intact fraction or a universal assay-success probability. Its interpretation also requires the specimen type and intended analysis: bulk <ribosomal RNA> features do not directly certify every individual <messenger RNA>.
Back to article page
