Search for non-coding RNA homologs and classify RNA families with Infernal covariance models (CMs) against Rfam. A CM scores both sequence and consensus secondary-structure conservation jointly, giving higher sensitivity than BLAST or profile HMMs for STRUCTURED RNAs whose sequence has diverged but whose structure is conserved. For RNAs without conserved structure (lncRNAs, mature miRNAs), a CM offers no advantage and a faster sequence method is correct.
# Infernal
conda install -c bioconda infernal
# Rfam database (pre-calibrated: press it, never recalibrate it)
wget https://ftp.ebi.ac.uk/pub/databases/Rfam/CURRENT/Rfam.cm.gz
gunzip Rfam.cm.gz && cmpress Rfam.cm
wget https://ftp.ebi.ac.uk/pub/databases/Rfam/CURRENT/Rfam.clanin
# Family-specialized tools (often the better choice for a known class)
conda install -c bioconda trnascan-se barrnap
# Python dependencies
pip install biopython pandas
Tell your AI agent what you want to do: - "Scan my transcripts against Rfam to classify ncRNA families" - "Find rRNAs, tRNAs, and snoRNAs in my genome" - "Build a covariance model for my novel RNA family" - "Should I use a covariance model or BLAST for this RNA?" - "Find all tRNAs in my genome assembly"
"I have non-coding transcripts. Scan them against Rfam and classify by family, resolving clan overlaps."
"Search this single RNA against Rfam and tell me what it is and whether the structure matches."
"Annotate ncRNA families in my bacterial genome using the documented Rfam pipeline."
"Find all tRNAs in my genome assembly with the right tool."
"My RNA is a lncRNA with no obvious conserved structure. Is a covariance model worth it?"
"I have a Stockholm alignment of a novel RNA family with a consensus structure. Validate the covariation, then build and search a CM."
"Parse the Infernal --fmt 2 output, deoverlap clans, and summarize family assignments."
--cut_ga for Rfam; the curated per-family bit-score cutoffs are database-size-independent. A flat -E 1e-5 reintroduces the DB-size dependence GA was built to avoid.-Z <Mb> for reproducibility, and prefer bit-score thresholds for comparison.--fmt 2 --clanin then drop hits marked = (grep -v ' = '); this is the documented Rfam deoverlap. --oclan is a valid alternative but not the documented path.cmcalibrate on any custom CM before trusting E-values (or threshold on bit score). Rfam.cm is already calibrated; only cmpress it.#=GC SS_cons; a structure-free alignment yields only an HMM, and the structure should pass R-scape covariation before you build.