BBTools Resources

Essential for metatranscriptome research.

The Problem: Raw metatranscriptome samples can contain 90-99% rRNA. This bloats files, wastes sequencing capacity, and can cause RNA-seq assemblers to crash. While wetlab ribodepletion kits remove the majority of rRNA particles, 50% of the remaining reads can still be rRNA.

Why Alignment Won't Work: Metatranscriptomes do not have known ribo sequences allowing alignment-based removal. Digital ribodepletion requires a different approach.

The Solution: riboKmers.fa.gz was designed using iterative passes of Silva through KCompress to find a minimal covering set of 31-mers. This captures >99.9% of 2x150bp read pairs synthetically generated from Silva SSU+LSU ribosomal sequences by prioritizing only the k-mers that, at each pass, capture the most reads.

Why Small is Better: At only 9MB, riboKmers minimizes the risk of coincidental sequence collisions with coding DNA while maintaining >99.9% rRNA capture efficiency.

Usage: Use with BBDuk for digital ribodepletion: bbduk.sh in=reads.fq out=clean.fq ref=riboKmers.fa.gz k=31

Created using: bbmap/pipelines/makeRiboKmers.sh

A new reference standard for prokaryotic SSU sequences.

The Problem with Existing Databases: Public SSU databases contain massive redundancy - the same organism can have 30+ sequences ranging from partial fragments to N-filled sequences. Which one represents the species? Aligning queries against highly redundant, incomplete databases containing partial and low-quality sequences is very slow, and yields inconclusive results.

The Canonical Selection Solution: Each organism's 16S sequences are processed through a sophisticated algorithm:

• All sequences aligned to create weighted traversal graph (HMM-like)
• Consensus sequence generated from most probable path
• Best real biological sequence selected using weighted scoring
• Naturally favors complete, high-quality sequences while avoiding pseudogenes and fragments

The Result: One high-quality, complete 16S sequence per organism. Dramatically more compact than Silva 138 - at 30 MB versus 688 MB - while covering 76,225 NCBI TaxIDs instead of 48,000. BBDB combines curated sequences from Silva with gene-called sequences from RefSeq genomes; 37% of sequences come from RefSeq.

Quality Features: No N's. No U's. TaxID-labeled headers. One representative sequence per organism. Tax-sorted for optimal gzip compression.

Impact: Dramatically fewer alignments required, minimal disk space, zero ambiguity about which sequence to use. Transforms taxonomic classification from navigating redundancy to obtaining direct, conclusive answers.

Eukaryotic SSU sequences, cleanly separated from organellar 16S.

The Eukaryote Problem: Both 16S and 18S are small subunit (SSU) rRNA. Eukaryotes have their own nuclear 18S, plus mitochondrial 16S. Plants add chloroplast 16S. Which represents the species? Mixed databases make finding the true eukaryotic SSU nearly impossible.

The Solution: Split databases. Prokaryotic 16S goes in one file, eukaryotic 18S in another. Organellar 16S sequences are excluded from the eukaryotic database. When both 16S and 18S are found for the same organism during gene-calling, the 16S is discarded and the 18S is considered representative.

The Result: One high-quality, complete 18S sequence per eukaryotic organism. Same canonical selection process as the prokaryotic database - consensus generation followed by best-sequence selection using weighted HMM-like scoring.

Quality Features: No N's. No U's. TaxID-labeled headers. Nuclear 18S only (no mito/chloroplast contamination). Tax-sorted for optimal gzip compression.

Usage: Integrated into BBTools programs like SendSketch and QuickClade for taxonomic classification. Provides exact SSU ANI (Average Nucleotide Identity) calculations via glocal alignment, complementing pentamer-frequency methods.