STREME writes its output to files in a directory named streme_out, which it creates if necessary. You can change the output directory using the --o or --oc options. The directory will contain:

• streme.html - an HTML file that provides the results in an interactive, human-readable format
• streme.txt - a text file containing the motifs discovered by STREME in MEME format
• sequences.tsv - a TSV (tab-separated values)
file that lists the true- and false-positive sequences identified by STREME for each motif
• sites.tsv - a TSV (tab-separated values)
file that lists the positions and scores of all sites of each discovered motif
• streme.xml - an XML file that provides the results in a format designed for machine processing

STREME gives each motif it discovers a name, which is a consensus sequence that approximately describes the motif, prefixed by a number from 1 to the number of motifs found, e.g., "1-CCACYAGT". Motif names are intended to be mnemonic only, and are not intended to be used for searching sequences for matches; to search for matches to STREME motifs, use a STREME output file as input to a motif scanning program such as FIMO and MAST.

Note: All options may be preceded by a single dash (-) instead of a double dash (--) if desired.

STREME searches for motifs by iterating the following five steps to until the selected stopping criterion is met. The stopping criteria are described in the "Number of Motifs" section, above.

1. Suffix Tree Creation.

   STREME builds a generalized suffix tree that includes both the primary and control sequences (but not the hold-out set sequences). If the alphabet is complementable, STREME adds the reverse complement of each primary and control sequence to the tree as well. For the first round of STREME, the input sequences are used to build the tree. For subsequent rounds, the sequences from the last step (Step 5, below), with previous motifs erased, are used.

2. Seed Word Evaluation.

   For each word of length 3 to the maximum motif width in the primary sequences, STREME uses the tree to efficiently count the numbers of exact matches to the word in the primary and control sequences, and computes its enrichment p-value using the chosen objective function. For each of the neval (default=25) most signficant words of each width, STREME then uses the tree to count the number of approximate matches to it, and computes its enrichment p-value. (Note: With the Differential Enrichment objective function, STREME will use the Binomial test instead of Fisher's exact test if the primary and control sequences have different average lengths. With the Central Distance objective function, STREME computes the cumulative Bates distribution of the average distances of the seed word from the centers of the sequences.)

3. Motif Refinement.

   For each word of length 3 to the maximum motif width in the primary sequences, STREME uses the tree to efficiently count the numbers of exact matches to the word in the primary and control sequences, and computes its enrichment p-value using the chosen objective function. For each of the neval (default=25) most signficant words of each width, STREME then uses the tree to count the number of approximate matches to it, and computes its enrichment p-value.

   STREME converts each of the nref (default=4) most significant seed words into a motif, and iteratively refines each motif, selecting the motif that best discriminates the primary sequences from the control sequences. At each iteration of refinement, the current motif and the (k-1)-order background are used with the suffix tree to efficiently find the best site in each sequence. The primary and control sequences are then sorted by the log-likelihood score of their best site, and the score threshold that optimizes the p-value of the statistical test (which depends on the chosen objective function) is found. The iteration ends by estimating a new version of the motif from the single best site in each primary sequence whose score is above the optimal threshold. This new motif is used in the next refinement iteration. Refinement stops when the p-value fails to improve, or when niter (default=20) iterations have been performed.

4. Motif Significance Computation.

   STREME computes the statistical significance of the of the motif by using the motif and the optimal discriminative score threshold (based on the primary and control sequences) to classify the hold-out set sequences, and then applying the statistical test (Fisher's exact test, Binomial test, or the cumulative Bates distribution) to the classification. Classification is based on the best match to the motif in each sequence (on either strand when the alphabet is complementable). If the primary or control hold-out set would contain fewer than 5 sequences, STREME does not create a hold-out set, and the value of the statistical test will not be an accurate measure of the statistical significance of the motif. In that case STREME will label the test output as a 'score' rather than as a p-value. If there is a hold-out set, STREME also reports an E-value that accounts for the number of motifs tested by STREME. The E-value is the p-value multiplied by the number of motifs reported by STREME, and is an estimate of how many equally enriched (or better) motifs would be found in shuffled versions of the primary sequences.

5. Motif Erasing.

   STREME "erases" each of the sites of the best motif from both the primary and control sequences by converting the sites to the separator character. Only the positions in the site where the letter has a positive likelihood ratio are erased to allow some overlap of sites of different motifs.

The running time and memory usage of STREME depends on the total size of the sequences in its input, the length of the sequences, the alphabet of the sequences and the minimum and maximum motif widths. The following tables show the running times and memory usage for STREME on on different size datasets (of random sequences), as a function of the length of the sequences, on a 3.2 GHz Intel Core i7 processor with 16GB of memory. STREME was run using a single thread, and the motif width was the maximum allowed range: minimum = 5, maximum = 30. The total sizes of the sequences in the datasets from 100,000 (1 x 10⁵) to 20,000,000 (2 x 10⁷).

With very short sequences (length = 5), STREME is extremely fast, processing 4,000,000 DNA sequences in less than 100 seconds. The running time increases with sequence length, reaching a maximum for sequences around 30 long. With very long sequences, STREME is runs more quickly, with length 10,000 sequences taking approximately the same time per motif discovered as length 10 sequences. STREME runs about twice as slow on DNA sequences as on RNA sequences because STREME treats DNA sequences as double-stranded and RNA sequences as single-stranded. Because of the larger alphabet size, STREME runs about 5 times more slowly on protein sequences than on RNA sequences, which is the same factor as the ratio of the alphabet sizes (20/4). The running time of STREME with sequences over a custom alphabet will similarly be (roughly) proportional to the number of letters in the custom alphabet.

STREME Running Time (seconds / motif)
DNA	RNA	Protein
STREME Virtual Memory Size (bytes)
DNA	RNA	Protein