SuperBLAST
About this platform

Standalone BLAST+, without the command line.

SuperBLAST (also referred to as BLAST Hub) exists because the difficulty most biologists have with BLAST is not the alignment science. It is everything around it — getting a hundred sequences in, keeping each result attached to the file it came from, and getting the whole set back out in a form the next step can use.

The gap this fills

The public web service

Authoritative and free, but it takes one query at a time, queues behind everyone else, and limits file and sequence size. A folder of a hundred reads means a hundred submissions and a hundred result pages to reconcile by hand.

Standalone BLAST+

No queue and no limits, but it asks the researcher to install the suite, build databases with makeblastdb, and write a script to loop over files. That is a fair amount of command-line competence to demand of someone whose expertise is at the bench.

Both routes fail the same person, and they fail for the same reason: the obstacle is data handling, not computation. SuperBLAST runs the genuine, unmodified BLAST+ binaries on managed infrastructure and returns one organised result set per submission. It is a front end to BLAST+ — not a reimplementation, and not a queue in front of the public NCBI service.

How it was built

Three approaches were developed and compared. The third is the one that worked.

01

BLAST inside a spreadsheet

The first attempt kept the biologist in the tool they already use, driving BLAST from Microsoft Excel. It broke on bulk work: the spreadsheet is a poor place to hold sequence data, and nothing about the approach removed the need for a local BLAST+ installation underneath it.

02

Standalone BLAST+ on an AWS virtual machine

The second attempt installed the full BLAST+ suite on a rented virtual machine. It worked, and it proved the engine was the easy part — but it left a server to provision, patch and pay for whether or not anyone was searching, and it still needed a command line to drive it.

03

A managed interface on Google Cloud

The third put the unmodified BLAST+ binaries and their databases inside a container image, ran it on managed infrastructure that scales with demand, and put a browser in front of it. No installation, no server to maintain, no command line — and the same engine underneath. This is the platform you are using.

Using it

01

Choose the program first

The program decides which molecule your query has to be and which databases you can search. blastn, blastx and tblastx take a nucleotide query; blastp and tblastn take protein. Setting this before you upload saves a rejected file.

02

Add your FASTA files

Drag in up to 100 files, 10 MB each and 50 MB per submission. Headers and residue alphabets are checked in the browser, so a malformed or wrong-molecule file is refused immediately with the reason — not halfway through a run.

03

Set the search parameters

Pick a reference database of the right molecule, a search task, an E-value threshold and a hit limit. The defaults are sensible for identification work: megablast and E = 0.01.

04

Read the results

Sort and filter by identity, coverage, E-value, bit score, organism or accession. Open any hit to see the pairwise alignment for each HSP with matches, mismatches and gaps marked.

05

Export

A multi-sheet Excel workbook with one sheet per query, a landscape PDF report, or the raw JSON and CSV. Results stay in your history for one year.

The five search programs, and the molecule each requires.
ProgramQueryDatabaseTranslation applied
blastnNucleotideNucleotideNone
blastpProteinProteinNone
blastxNucleotideProteinQuery, six reading frames
tblastnProteinNucleotideDatabase, six reading frames
tblastxNucleotideNucleotideBoth, six reading frames each

Reference databases

Every figure below was measured by blastdbcmd -info inside the production image. None of it is estimated.

Nucleotide and protein sets are provided for each organism, so every program has a compatible target. Letters are bases for nucleotide databases, residues for protein.
DatabaseOrganismAssemblyMoleculeSequencesLetters
drosophilaDrosophila melanogasterGCF_000001215.4RNA34,52692,449,215
drosophila_genomeDrosophila melanogasterGCF_000001215.4Genomic DNA1,870143,726,002
drosophila_proteinDrosophila melanogasterGCF_000001215.4Protein30,80220,379,498
ecoliEscherichia coli K-12 MG1655GCF_000005845.2Genomic DNA14,641,652
ecoli_proteinEscherichia coli K-12 MG1655GCF_000005845.2Protein4,3001,330,036
yeast_genomeSaccharomyces cerevisiae S288CGCF_000146045.2Genomic DNA1712,157,105
yeastSaccharomyces cerevisiae S288CGCF_000146045.2RNA6,1388,873,817
yeast_proteinSaccharomyces cerevisiae S288CGCF_000146045.2Protein6,0212,933,360
sarscov2SARS-CoV-2NC_045512.2Genomic RNA129,903
sarscov2_proteinSARS-CoV-2GCF_009858895.2Protein1214,149
virusesSARS-CoV-2 and HIV-1NC_045512.2, NC_001802.1Genomic RNA239,084

Each database is built with makeblastdb -dbtype {nucl|prot} -parse_seqids -taxid and the image carries NCBI’s taxonomy tables, so every hit resolves to a scientific name and a taxonomy identifier. The source URL, SHA-256 checksum, record count and index size are recorded inside the image for every database. The build fails if any database cannot be opened byblastdbcmd or does not resolve an organism name, and one search per program is run before the image ships — so a database that cannot be opened or searched cannot reach production.

How accuracy is checked

Every validation query is extracted from the database under test usingblastdbcmd, so no sequence is invented and each expected result is independently checkable.

  • Positive controls. A 600-base region of D. melanogasterchromosome 3R returns 100.000% identity over 600/600 positions, zero gaps, E = 0.0 at 1109 bits. The SARS-CoV-2 spike region behaves the same way.
  • Negative controls. A SARS-CoV-2 fragment searched against theD. melanogaster genome returns no hits at E < 1e-5, and anE. coli fragment against SARS-CoV-2 likewise.
  • A biological check. An E. coli 16S rRNA sequence searched against the E. coli genome returns exactly seven hits, at the chromosomal coordinates of the seven rrn operons, with rrnG and rrnDcorrectly reported on the minus strand. Seven is the number the biology predicts, and it is the strongest single piece of evidence that the pipeline is behaving.
  • Structural checks. A script in the repository asserts the JSON-15 structure and independently recomputes identity, coverage and gap openings from the alignment strings rather than trusting the reported fields.

What it does not do

Stated plainly rather than buried, because knowing the boundary is part of using a tool correctly.

  • The AI interpretation is not reproducible. Generation is non-deterministic and the model is a moving alias the provider repoints without notice. It has not been evaluated against expert annotation. Treat it as an exploratory aid, verify every claim against the alignments, and do not cite it. The BLAST results themselves are reproducible — engine version, assembly accessions and checksums are all pinned.
  • There is no published benchmark against standalone BLAST+ or the NCBI web service on this version.
  • The human GRCh38 database is disabled. Indexing it needs more scratch space than the deployment provides, so it is reported as unavailable with that reason rather than offered and failing.
  • Databases are a fixed set. You cannot yet supply your own.
  • Cancellation is best-effort when a search is running on a different server instance from the one receiving the request. The response says so explicitly rather than claiming a process was terminated when it was not.
  • Results are kept for one year and then removed automatically. Export anything you need to keep.

Where this goes next

This is a working prototype that validates an approach, not a finished product. The boundaries above are scope, not oversights — and each one points at a direction.

Beyond one hundred files

The current per-submission limit is a deliberate bound on how long a single job can run, not a limit of the engine. Moving execution to a worker pool with a job queue would let a submission scale to thousands of sequences and make cancellation reliable across instances at the same time.

User-supplied databases

Every database here is pinned and pre-indexed, which is what makes results reproducible — but it means you cannot search your own reference set. Allowing an uploaded FASTA to be indexed per user, with the same provenance recording, is the most requested extension of the idea.

NCBI E-utilities integration

Retrieving records by accession through Entrez, and linking hits to the literature, was attempted during the earlier virtual-machine work and remains the natural next step. It would let a result move from "what is this sequence" to "what is already known about it" without leaving the page.

More reference organisms

The set covers four model organisms plus a viral pair. Larger assemblies — human GRCh38 among them — need more scratch space than the current deployment provides for indexing, which is a provisioning question rather than a design one.

Citing this work

If SuperBLAST contributed to work you publish, please cite all three: the platform, the engine it runs, and the reference assemblies you searched against.

The platform

Vipin. SuperBLAST (BLAST Hub): bulk sequence search on managed NCBI BLAST+ infrastructure. Zenodo. https://doi.org/10.5281/zenodo.22688703

This is the concept DOI and always resolves to the most recent archived version. Machine-readable metadata is in CITATION.cff in the repository. ORCID 0009-0005-2924-7095.

The search engine

Camacho C, Coulouris G, Avagyan V, Ma N, Papadopoulos J, Bealer K, Madden TL. BLAST+: architecture and applications. BMC Bioinformatics 2009;10:421. https://doi.org/10.1186/1471-2105-10-421

SuperBLAST orchestrates NCBI BLAST+ 2.17.0 without modifying it. The algorithm and the statistics are theirs, and the original BLAST papers of Altschul and colleagues (1990, 1997) should be cited where you describe the method itself.

The reference data

All sequence data comes from NCBI RefSeq. Cite the assembly accession for each database you searched — they are listed in the table above, and the full source URLs and SHA-256 checksums are recorded in docs/DATABASES.md in the repository. Many journals ask that datasets be tagged [dataset] in the reference list.

Open source

The complete source is public under the MIT licence, including the container definitions, the database build scripts and the validation harness. Every release is archived on Zenodo with a permanent DOI.

github.com/vipin839/superblast

Origin and acknowledgements

This work began as an MSc dissertation in Biotechnology at Guru Jambheshwar University of Science and Technology, Hisar, supervised by Dr. Sapna Grewal.

Built on NCBI BLAST+, a United States Government Work in the public domain. BLAST® is a registered trademark of the National Library of Medicine. Reference data from NCBI RefSeq remains subject to NCBI’s data usage policies.

Try it on your own sequences

There is nothing to install and no cluster to provision.