Table 3. A comparison of the meme, hmmer, and AF methods for motif building.

Action	meme	hmmbuild	AF method
Input	An unaligned set of sequences. Motif region (if any) in each sequence may be unknown.	An aligned set of sequences. Nonhomologous sites may be removed prior to model building.	An aligned set of sequences. Entropy is used to differentiate con served from nonconserved sites.
Output	A set of PSSMs, one for each motif found by the algorithm.	Markov transition matrix specific to hmmer model.	A motif pattern that is mathematically very similar to a thresholded pssm.
Interpretability of the output	Not readily interpretable unless entries are thresholded or compared statistically.	Relatively uninterpretable. hmm is a nonconstructive statistical null hypothesis.	A readily interpretable motif pattern.
Strengths and weaknesses	Does not need an initial alignment to find or create motifs.	Requires initial sequence alignment.	Requires an initial sequence alignment.
	Search algorithm gives an estimate of how well the motif fits a test sequence.	Picking conserved regions to train model is subjective.	A priori biological knowledge can be included.
		Search algorithm gives an estimate of how well the motif fits a test sequence.	Mismatch count is correlated with probability of motif family membership.

PSSM, position-specific scoring matrix; thresholding refers to mapping if x < ε and otherwise.