Vaccine predictions
MHCflurry predicts binding and presentation. A Naive Bayes baseline evaluates TESLA candidates. The HCC1395 composite is a separate weighted rule, not that classifier.
Inspect vaccine ML evidence ↗RESEARCH WORKFLOW OVERVIEW
One platform.
Two research paths.
Vaccine candidate prioritization and ATF3 protein selectivity—mapped step by step, with the evidence, computation and remaining gaps in view.
SHARED PURPOSE
Make the next research decision traceable.
Both paths organize candidates, scores and evidence. Their biological questions and score meanings remain distinct. A common interface is not proof of a combined treatment.
THE WORKFLOW MAP
Follow the inputs. Inspect the outputs.
Open any step for its method, limitations and a link to the actual workspace.
Live computationStored data / resultBrowser input / exportNot connected
Separate evidence sources. TESLA is a public benchmark; HCC1395 is a separate public sample; ATF3 uses reference protein sequences. They are not a matched patient dataset.
WHERE AI FITS
Model predictions and rules play different roles.
Protein structures
AlphaFold2-Multimer generated the supplied structures. The browser displays them; editing a sequence does not run a new prediction.
Review stored structures ↗ATF3 selectivity screen
Register detection, contact counting and constrained enumeration run on the visitor's device. No learned ATF3 affinity or efficacy model is deployed.
Run the actual analysis ↗THE CONNECTION ROADMAP
Build the bridge. Don't assume it exists.
These are proposed integration and validation steps—not running modules.
A critical scientific boundary
Changing an ATF3 variant's self/partner contact scores does not establish inhibition of endogenous ATF3. Nor does it demonstrate vaccine synergy. The intended intervention and functional evidence must be established separately.
CONTINUE WITH THE EXISTING TOOLS