Table 1 Space and time requirements for two gene finders Two gene finders, the 31-state DSP gene finder TIGRscan, and the 6-state PSA gene finder GlimmerHMM, were run on a 922 Kb sequence. The DSP gene finder used raw probabilities and the PSA gene finder used log-likelihood ratios. The DSP implementation required less memory, both in total and per state, than the PSA implementation. Although the PSA implementation required less total time, the DSP implementation required the same amount of time per state, so that for a given gene finder with a fixed number of states, DSP decoding can be expected to be fully as fast as PSA decoding.

From: Efficient decoding algorithms for generalized hidden Markov model gene finders