Tree-of-life scale genomics
Tree-of-life–scale homology search built on object storage. Fast, accurate, 10× cheaper, and infinitely scalable.
Rafflesia lets you run Alphafold 100x faster.
AI agents are ushering a new era in biology, reasoning across billions of sequences to surface connections, accelerate discovery and invent new medicines.
Rafflesia is building the fastest and most scalable cloud infrastructure for the AI era.
We perform the fastest prediction for Alphafold, ESMFold and many more, with a simple, plug-and-play, agent-ready API.
If you want to deploy rafflesia on your own private cloud, contact us.
- Read query142 aa · hemoglobin subunit alpha
- Alignaffine Smith-Waterman, BLOSUM62
- Rankby alignment score
Scale
Our homology infrastructure separates storage and compute. Sequences are indexed on cheap object storage, and expensive Smith-Waterman dynamic programming alignment is performed only on the subset of data that's the most relevant. This allows us to achieve tree-of-life scale protein alignment in seconds, with an accuracy far superior to BLAST, MMseqs2 and DIAMOND.
The approach was pioneered in the 2010s by the engineers behind Google's Dremel and BigQuery, to run SQL queries over petabytes of data. It now powers the most scalable databases on the web: Snowflake, Databricks, ClickHouse, and more.
LATENCY (MS)
RECALL (%)
FAQs
Rafflesia runs homology search over the protein universe: give it a sequence or a structure and it returns the related ones, aligned, with coordinates, coverage, and a significance estimate. It provides deterministic biological primitives with the evidence behind every result, leaving the project-specific interpretation to you.
Those tools match sequences letter by letter. Once two proteins share less than about a quarter of their residues, the letters stop agreeing and those tools go quiet, even though the proteins are still related. Rafflesia finds those pairs. Each one comes back as a full alignment: where the match starts and ends, how much of each sequence it covers, and how likely it is to be chance.
The other difference is scale. Their index has to fit on the machine running the query. Ours does not, so a bigger database does not mean bigger hardware.
UniProt for sequences, and the PDB plus AlphaFold DB for structures, which is over 200 million of them. Every search names the release it ran against, so a result stays reproducible after the underlying database moves on.
Yes. Private collections are indexed the same way the public ones are and searched with the same primitives. Get in touch and we will size it with you.
The aligned pair with query and target coordinates, the alignment itself, coverage, the score, and a significance estimate calibrated against the space that was actually searched. Everything is machine-readable and carries the release it came from, because a hit you cannot trace is not evidence.
Yes. The same query against the same release returns the same answer, and no language model sits anywhere in the path. Nothing is sampled, nothing is guessed, and nothing summarises the result for you.
An HTTP API, the rafflesia CLI, and generated SDKs that all speak the same contract, because they are all generated from it. Searches run asynchronously and batch cleanly, so ten thousand queries are one job rather than ten thousand of them.
Tree-of-life scale searches finish in seconds. A much larger corpus does not mean a larger deployment, so growth in the database is not growth in what you have to run.
Install the CLI, or open the dashboard and run a search against a public release. Neither needs infrastructure from you.
Rafflesia homology
Search biology at corpus scale.
Bring a corpus, a homology-search workload, or a recall-versus-cost problem. We’ll help you compile it into a release and serve it.