TL;DR
Funnel a real wet-lab NGS pool from about 10^6 sequences to a ranked shortlist of about 10^2 candidates.
Use it / Skip it
The lab supplies diversity; the platform supplies staged filtering, structure/affinity scoring, epitope/specificity checks, and an auditable shortlist.
Use when
You have an enriched sequence pool and need a bench-usable shortlist for expression or validation.
Don't use for
Do not use it when there is no sequence pool yet. Upload sequences inline or through NGS Pool first.
Inputs
- Target
- A configured target such as BCMA or another target with readiness metadata.
- Sequence pool
- An ingested pool or an inline FASTA/CSV upload.
- Shortlist size
- Default around 100; tune for downstream lab capacity.
- Options
- Spike known positives, high-precision docking, and structure-based scoring where available.
Outputs
- Stage counts
- Pool, filter, fold/affinity, epitope/specificity, composite rank, shortlist.
- Recent runs
- Persisted jobs with status and selectable detail.
- Shortlist
- Ranked candidates with evidence suitable for the next wet-lab step.
Walkthrough
1. Pick a target first
Pick a target first. The pool selector is target-aware so you do not launch runs that cannot resolve.
2. Choose an existing pool or upload FASTA/CSV inline
Choose an existing pool or upload FASTA/CSV inline. Ingested pools become immediately selectable.
3. Set shortlist size and optional gates
Set shortlist size and optional gates. Use anchor spiking when the target has known positives.
4. Run the funnel and let polling finish
Run the funnel and let polling finish. Read stage counts before trusting the final top list.
5. Open the validation report when available
Open the validation report when available. Known positives recovering is the trust check before unlabeled production pools.
Under the hood
- Uses semi-de-novo API helpers in lib/api for pool upload, run submission, job polling, and job list refresh.
- The implemented funnel separates cheap sequence filters from expensive structure-based scoring.
- Validation report records DLL3 known-positive recovery and BCMA production-run behavior.
Worked example
Pitfalls
- If the pool dropdown is empty, the target probably has no compatible pool yet; upload sequences first.
- Docking all 10^6 clones is intentionally avoided; the first stages are cheap sequence filters.
- Predicted affinity is not measured Kd. Wet-lab KD and specificity remain the truth source.