NeuroFold: an RNA secondary structure prediction system using a Hopfield neural network

Yutaka Akiyama (akiyama@kuicr.kyoto-u.ac.jp)
Minoru Kanehisa

Institute for Chemical Research, Kyoto University

Abstract

We have developed a quick algorithm for RNA secondary structure prediction which employs a combinatorial optimization technique based on the Hopfield neural network theory. The detail of our algorithm was presented at the last Genome Informatics Workshop in 1991. Briefly, the algorithm consists of four independent stages. At the first stage all possible stacking region candidates are listed via a string matching process along a given RNA sequence. (Several different search strategies can be designated by options.) Then the second module executes selective deletions of redundant candidates. And the third, a Hopfield neural network module performs the central task of constructing the most feasible combination set from proposed candidates. The technique of dynamic modification of neural I/O function is used in order to escape from local minima problem. The last stage is to build an output display from the selected set of stack region candidates.

In this workstation session, we would like to demonstrate performance of the "NeuroFold" system which is an X11 window-based implementation of the proposed algorithm. The algorithm itself has also been enhanced in many aspects: for example, now pseudoknots can be effectively predicted in the NeuroFold system.