Question

My algorithm predicts for a given amino acid sequence several measurable properties.
Now given a table with the real, measured values and my values how do i compute how
well my algorithm is performing?

Example:

Sequence Property1 Prediction1 
seq1     3l.4      35.0         |Prediction matches well
seq2     23.4      79.9         |Prediction was off
seq3     20.0      20.3         |Prediction as super

==> The algorithm performed well.

Is there a way to quantify this?

Answer 1

Yes, create an index of quality.

The simplest would be least squares – calculate the difference between a result and the measured value, square it, and sum the squares. Divide by the number of values. The square root of then gives you the standard error norm.

However, that assumes that it is equally important to get the right answer for all of the measurements. If some are more important than others, then you should weight them in the sum.

Try asking yourself the question of what would be twice as bad as your answer above. Double all the errors?