Version Compatibility

Reference examples tested with: pandas 2.2+

Before using code patterns, verify installed versions match. If versions differ:

• Python: pip show <package> then help(module.function) to check signatures
• CLI: <tool> --version then <tool> --help to confirm flags

If code throws ImportError, AttributeError, or TypeError, introspect the installed package and adapt the example to match the actual API rather than retrying.

Non-Coding RNA Annotation

"Find non-coding RNAs in my genome" → Scan a genome assembly for rRNAs, tRNAs, snoRNAs, and other ncRNA families using covariance model search and specialized detectors.

• CLI: cmscan --rfam --tblout hits.tbl Rfam.cm assembly.fa (Infernal), tRNAscan-SE -o trnas.txt assembly.fa

Identify and annotate non-coding RNAs in genome assemblies using Infernal (general ncRNAs via Rfam covariance models) and tRNAscan-SE (specialized tRNA detection).

Infernal / cmscan

Infernal uses covariance models (CMs) from Rfam to identify ncRNA families by both sequence and secondary structure similarity.

Rfam Database Setup

# Download Rfam covariance models (~600 MB compressed)
wget https://ftp.ebi.ac.uk/pub/databases/Rfam/CURRENT/Rfam.cm.gz
gunzip Rfam.cm.gz

# Press the CM database (required for cmscan)
cmpress Rfam.cm

# Download clan information (for overlap resolution)
wget https://ftp.ebi.ac.uk/pub/databases/Rfam/CURRENT/Rfam.clanin

Basic cmscan

# Search genome against all Rfam families
# --cut_ga: Use Rfam gathering threshold (recommended, family-specific)
# --rfam: Run in Rfam mode (skip slow alignment for large models)
# --nohmmonly: Use CM scoring only (more accurate, slower)
# --fmt 2: Output in table format
cmscan \
    --cut_ga \
    --rfam \
    --nohmmonly \
    --tblout ncrna_hits.tbl \
    --fmt 2 \
    --cpu 16 \
    --clanin Rfam.clanin \
    Rfam.cm \
    genome.fasta > ncrna_hits.out

Key Options

| Option | Description | |--------|-------------| | --cut_ga | Use Rfam gathering threshold (recommended) | | --rfam | Speed optimization for Rfam-scale searches | | --nohmmonly | Force CM-only scoring (more sensitive for structured RNAs) | | --clanin | Rfam clan file for resolving overlapping hits | | --tblout | Tabular output file | | --fmt 2 | Extended table format with accessions | | --cpu | CPU threads | | -E | E-value threshold (default: 10; --cut_ga is preferred) |

Convert cmscan Output to GFF3

# Using Infernal's built-in conversion
grep -v '^#' ncrna_hits.tbl | \
awk 'BEGIN{OFS="\t"} {
    if ($10 == "+") strand = "+"; else strand = "-";
    if ($10 == "+") {start=$8; end=$9} else {start=$9; end=$8};
    print $4, "Infernal", "ncRNA", start, end, $17, strand, ".", "ID="$2";Name="$3";rfam_acc="$2";evalue="$17
}' > ncrna_infernal.gff3

E-value Guidelines

| Category | Typical E-value | Notes | |----------|----------------|-------| | High confidence | < 1e-10 | Unambiguous family assignment | | Moderate | 1e-10 to 1e-5 | Likely real, verify context | | Rfam GA threshold | family-specific | Recommended cutoff with --cut_ga | | Marginal | > 1e-3 | May be pseudogenes or degraded copies |

tRNAscan-SE

Specialized tRNA detector using covariance models. More sensitive and specific for tRNAs than Infernal alone.

Basic Usage

# Bacterial/archaeal mode
tRNAscan-SE -B -o trna_results.txt --gff trna.gff3 genome.fasta

# Eukaryotic mode
tRNAscan-SE -E -o trna_results.txt --gff trna.gff3 genome.fasta

# General mode (auto-detect domain)
tRNAscan-SE -G -o trna_results.txt --gff trna.gff3 genome.fasta

Key Options

| Option | Description | |--------|-------------| | -B | Bacterial mode | | -A | Archaeal mode | | -E | Eukaryotic mode | | -G | General (mixed/unknown) mode | | -o | Tabular output | | --gff | GFF3 output | | --detail | Detailed output with isotype info | | --thread | CPU threads | | -Q | Covariance model scoring only (slower, more accurate) |

Domain-Specific Models

| Domain | Flag | Notes | |--------|------|-------| | Bacteria | -B | Shorter introns, smaller genomes | | Archaea | -A | Split tRNAs, bulge-helix-bulge introns | | Eukaryota | -E | Longer introns, pseudogenes common | | Organelle | -O | Mitochondrial/chloroplast tRNAs | | Mitochondrial | -M | Mammalian mitochondrial-specific |

Expected tRNA Counts

| Organism Type | Expected tRNAs | Notes | |---------------|----------------|-------| | Bacteria | 30-90 | Fewer isotypes, some shared | | Archaea | 30-60 | Similar to bacteria | | Yeast | 275-400 | Many isodecoders | | Nematode | 600-900 | Gene family expansion | | Mammal | 400-600 | Many pseudogenes | | Plant | 500-1,000+ | Large gene families |

barrnap (rRNA Detection)

barrnap predicts rRNA genes using HMM models. Note: barrnap has been unmaintained since 2018. Bakta handles rRNA detection internally for prokaryotes.

# Detect rRNAs
barrnap --kingdom bac genome.fasta > rrna.gff3

# Kingdoms: bac (bacteria), arc (archaea), euk (eukaryote), mito (mitochondria)
barrnap --kingdom euk --threads 4 genome.fasta > rrna.gff3

Combining ncRNA Annotations

Goal: Merge Infernal and tRNAscan-SE results into a unified ncRNA annotation, using the best tool for each RNA type.

Approach: Parse Infernal tabular output and classify hits by Rfam family name into broad ncRNA categories, parse tRNAscan-SE GFF for tRNA calls, then combine by dropping Infernal tRNA hits (tRNAscan-SE is more accurate for tRNAs) and keeping Infernal for all other ncRNA types.

import pandas as pd
from collections import defaultdict

def parse_infernal_tbl(tbl_file):
    '''Parse Infernal cmscan tabular output.'''
    hits = []
    with open(tbl_file) as f:
        for line in f:
            if line.startswith('#'):
                continue
            parts = line.split()
            if len(parts) < 18:
                continue
            strand = '+' if parts[9] == '+' else '-'
            start, end = (int(parts[7]), int(parts[8])) if strand == '+' else (int(parts[8]), int(parts[7]))
            hits.append({
                'target': parts[0],
                'rfam_acc': parts[1],
                'rfam_name': parts[2],
                'seqid': parts[3],
                'start': start,
                'end': end,
                'strand': strand,
                'evalue': float(parts[16]),
                'score': float(parts[15]),
                'rna_type': classify_rfam(parts[2]),
            })
    return pd.DataFrame(hits)

def classify_rfam(rfam_name):
    '''Classify Rfam family into broad ncRNA categories.'''
    name_lower = rfam_name.lower()
    if 'rrna' in name_lower or 'ssu' in name_lower or 'lsu' in name_lower:
        return 'rRNA'
    if 'trna' in name_lower:
        return 'tRNA'
    if 'snorna' in name_lower or 'snord' in name_lower or 'snora' in name_lower:
        return 'snoRNA'
    if 'mir' in name_lower:
        return 'miRNA'
    if 'riboswitch' in name_lower or 'thermoregulator' in name_lower:
        return 'riboswitch'
    if 'ires' in name_lower or 'leader' in name_lower:
        return 'cis-reg'
    return 'other_ncRNA'

def parse_trnascan_gff(gff_file):
    '''Parse tRNAscan-SE GFF3 output.'''
    trnas = []
    with open(gff_file) as f:
        for line in f:
            if line.startswith('#'):
                continue
            parts = line.strip().split('\t')
            if len(parts) < 9 or parts[2] != 'tRNA':
                continue
            attrs = dict(item.split('=') for item in parts[8].split(';') if '=' in item)
            trnas.append({
                'seqid': parts[0],
                'start': int(parts[3]),
                'end': int(parts[4]),
                'strand': parts[6],
                'isotype': attrs.get('isotype', 'unknown'),
                'anticodon': attrs.get('anticodon', 'unknown'),
                'score': float(parts[5]) if parts[5] != '.' else 0,
                'rna_type': 'tRNA',
            })
    return pd.DataFrame(trnas)

def combine_ncrna_annotations(infernal_tbl, trnascan_gff, output_gff):
    '''Combine Infernal and tRNAscan-SE results, preferring tRNAscan for tRNAs.'''
    infernal_df = parse_infernal_tbl(infernal_tbl)
    trna_df = parse_trnascan_gff(trnascan_gff)

    # Remove Infernal tRNA hits (tRNAscan-SE is more accurate for tRNAs)
    infernal_no_trna = infernal_df[infernal_df['rna_type'] != 'tRNA']

    summary = defaultdict(int)
    for rna_type in infernal_no_trna['rna_type'].unique():
        summary[rna_type] = len(infernal_no_trna[infernal_no_trna['rna_type'] == rna_type])
    summary['tRNA'] = len(trna_df)

    print('=== ncRNA Summary ===')
    for rna_type, count in sorted(summary.items()):
        print(f'  {rna_type}: {count}')
    print(f'  Total: {sum(summary.values())}')

    return infernal_no_trna, trna_df, summary

Troubleshooting

cmscan Is Slow

• Use --rfam flag for Rfam-mode optimizations
• Use --FZ 2 to further speed up (less sensitive)
• Split genome into chunks and run in parallel

Missing Expected ncRNAs

• Verify Rfam.cm is current version
• Check that cmpress was run successfully (look for .i1m, .i1p, .i1f, .i1i files)
• For mitochondrial/chloroplast ncRNAs, search organelle sequence separately

tRNAscan-SE Finds Too Many Pseudogenes

• Normal for eukaryotic genomes (especially mammals)
• Filter by score: high-confidence tRNAs typically score > 50 bits
• Use --detail flag to identify pseudogenes in output

Related Skills

• prokaryotic-annotation - Bakta includes ncRNA annotation
• eukaryotic-gene-prediction - Gene prediction excludes ncRNAs
• rna-structure/ncrna-search - Targeted ncRNA homology searches