Build local BLAST+ databases and run searches using the BLAST+ CLI. Encodes BLAST v5 vs v4 database format and taxonomy filtering, -task taxonomy (megablast/dc-megablast/blastn/blastn-short and the blastp variants), soft vs hard masking, thread saturation past ~16 threads, output-field semantics (pident vs qcovs vs qcovhsp), the max_target_seqs misuse (Shah 2019), the -parse_seqids requirement for downstream extraction, and the size/reproducibility trap of nt/nr.
conda install -c bioconda blast
blastn -version # 2.15+ expected
update_blastdb.pl --showall pretty | head
For large workflows (>100,000 queries), consider DIAMOND or MMseqs2 instead -- 100-10,000x faster at comparable sensitivity. See remote-homology.
"Build a protein BLAST database from reference_proteins.fasta. Use -blastdb_version 5 and -parse_seqids so blastdbcmd can extract sequences by accession later, and -hash_index for faster lookups."
"I'm doing cross-species DNA homology between mouse and human. Use -task dc-megablast (discontiguous, sensitive). Don't use default megablast -- that needs 28-nt exact-match seeds and misses cross-species hits."
"Download NCBI's curated set for cross-species homology. Use refseq_select_rna (~5 GB) via update_blastdb.pl, not nt (~250 GB). nt also isn't reproducible -- it changes daily."
"I have a 32-core box. Don't set -num_threads 32 -- BLAST is I/O bound past ~16 threads. Either cap at 16 threads, or split the query FASTA into 4 chunks and run 4 parallel blastn processes with -num_threads 8 each."
"Set -max_target_seqs 500 and post-filter to top 10 by bit-score. Don't set -max_target_seqs 10 -- it's an early-termination heuristic that biases what's returned (Shah et al. 2019 Bioinformatics 35:1613)."
"After blastp finishes, use blastdbcmd -db ref_prot_db -entry_batch hit_accessions.txt to pull the subject FASTA sequences for downstream analysis."
blastn -version.-parse_seqids + -hash_index.-task deliberately: megablast for >=95% identity, dc-megablast for cross-species, blastn for general, blastn-short for <50 nt.refseq_select over nt/nr for reproducibility unless explicitly needed otherwise.-num_threads between 8 and 16; split input for higher parallelism.-outfmt "6 qseqid sseqid pident length qcovs qcovhsp evalue bitscore staxids sscinames stitle" for analyzable tabular.-max_target_seqs 500+ and post-filter; cite Shah 2019 if the user asks why.-taxids / -taxidlist for taxonomy filtering (v5 only); fall back to entrez_query for v4 (remote).blastdbcmd only if DB was built with -parse_seqids.update_blastdb.pl --decompress overnight; sizes can exceed 100 GB.pident in -outfmt 6 is identity over the HSP, not the query; use qcovs (total query coverage) for the "coverage" most people mean.-blastdb_version 5 so -taxids works. v4 databases require the taxonomy4blast.sqlite3 companion file and don't support -taxidlist.remote-homology.ortholog-inference.subprocess for Python integration.