next up previous
Next: Bibliography Up: SPEP analysis Previous: Combining SignalNet with CleavageNet

Predicting the Cleavage Sites

Even though the accuracy obtained for the CleavageNet is remarkable, it is not easy to sort out the exact position of the cleavage site. For this reason, we used an idea similar to that previously used [10] which takes advantage of both type of neural networks. We then introduced a function
\begin{displaymath}
Comb(i)=\sqrt{dS(i)^2 \cdot O_C(i)} \; .
\end{displaymath} (9)

where $O_C(i)$ is the usual CleavageNet output for the residue position $i$ and $dS(i)$ is a kind of numerical derivative of the SignalNet. Considering the SignalNet output ($O_S$)
\begin{displaymath}
dS(i)=\frac{\sum_{k=1}^{w+a}O_S(i-k) - \sum_{k=0}^{w}O_S(i+k)}{w+a} \; .
\end{displaymath} (10)

where $w$ (usually set around 10) and $a$ (around 10) are used to take into account the N-terminal part with respect to the upstream positions. An example is reported in Fig. 1. It is evident that more than one picks are generated by CleavageNet, and the one centered in $i=31$ (the expected observation) is not the highest. However, taking into account the values of SignalNet, as describe by 9, $Comb$ assigns the correct location.

Figure 1: An example of the prediction of AMO_ECOLI using the SignalNet, CleavageNet and the Comb methods. Y-axis: method-outputs. X-axis: residue position along the chain.
\begin{figure}
<tex2html_file> ...



next up previous
Next: Bibliography Up: SPEP analysis Previous: Combining SignalNet with CleavageNet
2003-06-12