Encodage linéaire d'automates pondérés. Filtrage de motifs génomiques et application sur l'architecture prototype R-disk

Abstract

We show that the linear encoding scheme (or one-hot scheme) efficiently simulates weighted finite automata (WFA). Those automata carry weights on every transition and model substitution errors in proteic patterns. Automata with t transitions can be avantageously hardwired with 0(t) operators. This scheme solves pattern matching problems by feeding a pipeline with one character every clock cycle. Such automata are well suited for use in FPGA devices, especially within the R-disk prototype, a hardware architecture devoted to content-based searches inside non-indexed large databanks: data is filtered on-the-fly at the output of storage devices, using distributed and reconfigurable processing elements. This improves the speed of parsing genomic databanks.