- Problem
- Stargardt disease type 1 has no approved ASO therapy for this deep-intronic variant, and rational ASO design requires extensive sequence analysis, scoring and experimental validation.
- Solution
- A Python pipeline that scores candidate PMO ASOs, documents known issues openly and includes an adversarial review by six independent reviewers before any claim is made.
- My role
- Research lead
- Result
- A transparent research prototype with 224 passing tests, open methodology and disclosed limitations — no synthesized or tested ASOs.
Overview
A reproducible computational pipeline for designing candidate antisense oligonucleotides against a deep-intronic ABCA4 variant, built with full disclosure of its limitations.
Problem
Therapeutic oligonucleotide design is hard to validate without lab work, and computational papers often hide the gap between prediction and evidence.
My role
Research lead. I scoped the chemistry choice, commissioned the adversarial review and interpreted the clinical boundaries of the predictions.
Solution
A Python pipeline that scores PMO ASO candidates, reports evidence levels for every claim, and subjects itself to independent review before publication.
Technical approach
Sequence analysis, scoring heuristics, 224 automated tests, a public documentation vault and an adversarial review panel of six independent reviewers.
Challenges
Bridging computational prediction and therapeutic reality without experimental data, and keeping every claim tied to an explicit evidence level.
Results
The pipeline passed 224 tests and survived adversarial review. The verdict was 'reject with invitation to major resubmission' — exactly the honest outcome the project was designed to surface.
What I learned
That AI-assisted research is only credible when it invites scrutiny, and that the most useful output of a prototype can be a clear list of what would need to change for it to be trusted.
Technologies
- Python
- Bioinformatics
- PMO chemistry
- ASO design
- Testing
- Reproducible research