Table of Contents Searching for particular repeats Overview search_hmm Full usage filter_hmm Full usage Python API search_hmm Full usage filter_hmm Full usage Use GC3Pie to annotate your large sequence dataset. Running TRAL using vagrant The search package provides python modules and command line tools for identifying tandem repeats in sequence databases. Typically processing consists of first running search_hmm to identify instances of a query repeat in a sequence database. This is relatively slow (hours to days against Uniprot) but can be trivially parallelized to run on a cluster. After scoring the hmm against every sequence in the database, filter_hmm is used to identify significant results. The search module can be run directly: The query HMM should be in HMMR format. The hmmbuild tool may be helpful for generating HMM files from multiple alignments. The HMM should represent a single repeat. Circularly permuted edges are automatically inserted to permit multiple sequential repeats. The database should be in FASTA format. It may be gzip compressed. Results are produced in TSV format, and include Sequence identifier (header line from FASTA file) Sequence identifier (header line from FASTA file) Viterbi probability (probability that this sequence was emitted by the query HMM) Viterbi probability (probability that this sequence was emitted by the query HMM) Log-odds ratio (viterbi probability divided by the expected probability for a sequence of this length Log-odds ratio (viterbi probability divided by the expected probability for a sequence of this length HMM states (Maximum likelihood sequence of states through the HMM, as determined by the Viiterbi algorithm HMM states (Maximum likelihood sequence of states through the HMM, as determined by the Viiterbi algorithm

When running on a cluster, it may be useful to search just a subset of the database. For instance, here’s a simplified submission script using the -n and -s parameters in an array job: Choosing an appropriate log-odds ratio depends on the database size, amino acid content, HMM length, and HMM composition. To help calibrate this, the -r option is available to automatically shuffle the input sequences. The results then represent all negative hits, allowing the False Discovery Rate to be estimated for various thresholds in the filtering step. The results from search_hmm include both significant and insignificant hits. Use filter_hmm to filter the results based on length and log-odds criteria. Output . The tool can produce three kinds of output: -f FASTA : Subset of the input database corresponding to significant hits -f FASTA : Subset of the input database corresponding to significant hits -o TSV : Subset of the search_hmm results corresponding to significant hits -o TSV : Subset of the search_hmm results corresponding to significant hits -x TREKS : All significant repeats in T-REKS format. -x TREKS : All significant repeats in T-REKS format. The TREKS format is most useful for exporting results to other programs. Identifiers . By default, the identifier is extracted from the database header lines (e.g. >tr|id|description... ). For compatibility with other programs, the --preserve-header option can also be used to match sequences based on the full header. Thresholds . Thresholds can be set for either: -r Number of repeats (float). Repeats are computed based on the HMM match states. This means that partial repeats are counted as a fraction of the query length. -r Number of repeats (float). Repeats are computed based on the HMM match states. This means that partial repeats are counted as a fraction of the query length. -t Log-odds threshold (float). Results with log-odds less than this value are filtered out. -t Log-odds threshold (float). Results with log-odds less than this value are filtered out. Memory usage . Memory usage should be low if a single output is used. If several output options are selected then the significant hits are stored in memory. If this is problematic, filter_hmm can be run separately for each desired output. See also the Search code documentation