A new approach to computer-assisted gene recognition in higher eukaryo
te DNA is suggested, It allows one to use not only linear functions fo
r scoring structures, but all functions satisfying natural monotonicit
y conditions, The algorithm constructs the set of structures guarantee
d to contain an optimal structure for every function, So, it uncouples
the time-consuming step of generation of this set from the fast step
of structure scoring, thus making it simple to experiment with differe
nt functions, One particular scoring function, taking into account onl
y codon usage and positional nucleotide frequencies of the splicing si
tes, has been implemented in the Genome Recognition and Exon Assembly
Tool program, and has been tested on an independent sample of human ge
nes, yielding 88% sensitivity and 79% specificity.