Protein bioinformatics
Open sourcev0.1.1
Neural network-powered
bioinformatics.
Learned protein alignment, clustering and search. Explore a pair in the browser demo, or use Orikata Tools to analyze your own proteins and compare methods.
Run Hikoboshi in Orikata Tools.
One workspace for Hikoboshi and established sequence and structure methods. Choose an analysis, bring your inputs, and explore the results.
Try Hikoboshi
Pick two proteins
and see how they align.
Loading the example proteins…
Compare what the
model has learned.
Hikoboshi turns each residue into a learned representation. Comparing those representations reveals correspondences between proteins, even when their sequences differ.
The workspace above follows a pair from structure to embeddings, alignment and superposition. Bring your own structures, inspect individual residues, and take the results with you.
The algorithm
inside the model.
Orihime makes sequence alignment differentiable. That lets Hikoboshi’s encoders learn through the alignment itself.
Native model execution puts those learned comparisons into practical bioinformatics workflows. This work also became the starting point for Hataya.
See how Orihime worksUse Hikoboshi
From two proteins
to whole collections.
View Hikoboshi on GitHub About the name
Hikoboshi
- English guide
- hee-koh-boh-shee
- Japanese IPA
- [çi.ko.bo.ɕi]
彦星 · The cowherd celebrated at Tanabata, Japan’s star festival.
Hikoboshi (Altair) and Orihime (Vega) are the lovers in the Tanabata story, separated by the Milky Way and reunited once a year. In our software, they meet whenever we train a new network with Orihime and deploy it in Hikoboshi.
All meanings and word history