Figure 1

Comparison of 6 different network-based methods for essential amino acids prediction. For each method (Methods 1–6 and "Random <1%") the abscissa x is the proportion of amino acids predicted (by the appropriate centrality measurement) and y is the proportion of essential amino acids predicted (i.e. sensitivity). For example, x = 0.2 means that we select 20% of the amino acids. If we consider the closeness centrality, it means that we select 20% of the amino acids that have the largest closeness centrality. We can notice in this case that we select around 38% of the essential amino acids. The slide of the "Maximum value" curve is the proportion of amino acids that are essential. All the curves have to be under this curve. The better a method is, the closer to the "Maximum value" its associated curve is. A method is better than random if its curve is over the "Average value: y = x" curve. A method is worse than random if its curve is under the "Average value: y = x" curve. The calculations (of the curves 1–6) are made as an average over the five networks representing HIV-1 protease, TEM1 beta-lactamase, T4 lysozyme, Barnase and bacteriophage f1 gene V protein. The curve "Random <1%" only depends on the average number of amino acids of the proteins. There is a probability less than 1% that a random selection of the amino acids will produce a curve over the "Random <1%" curve.