AI Science Tracker

Deep learning designs abiotic miniproteins that improve antisense-oligonucleotide nuclear delivery

Massachusetts Institute of Technology; Broad Institute of MIT and Harvard; Harvard Medical School; Sarepta Therapeutics · 9 Aug 2021 · medicine

Discovery

Deep learning designs abiotic miniproteins that improve antisense-oligonucleotide nuclear delivery

A generator–predictor–optimizer deep-learning workflow trained on quantitative delivery assays was used to design abiotic Mach miniproteins. Twelve selected predicted constructs were synthesized/tested; the strongest library construct improved phosphorodiamidate morpholino oligomer (PMO) delivery nearly 20-fold over unconjugated PMO in the standardized HeLa assay, while the authors describe the best designs as up to 50-fold more effective than prior polypeptide-based variants. PMO–Mach3/4 also produced dose-dependent reporter-protein restoration in skeletal muscle, diaphragm and heart in transgenic mice after a single intravenous dose.

Technology: Deep-learning generator–predictor–optimizer workflow: nested-LSTM sequence generator, CNN activity predictor on residue chemical fingerprints, and evolutionary optimization

Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.

Result, AI contribution and limits

Evidence context (descriptions follow the cited sources; test type is a catalog classification):

  • Publication status: Peer-reviewed paper
  • Test type: direct test
  • Tested: A generator–predictor–optimizer deep-learning workflow trained on quantitative delivery assays was used to design abiotic Mach miniproteins.
  • Still unknown: The paper reports improved delivery in a standardized HeLa reporter assay and transgenic mouse reporter model, not therapeutic correction of a disease. Only selected designs were synthesized (12 from hundreds of predicted sequences); mouse experiments used an EGFP reporter transgene and short seven-day follow-up. The article separately states the library's highest construct was nearly 20-fold over PMO alone and characterizes the designs as 50-fold better than previously known polypeptide variants; these denominators are not interchangeable. Safety evidence is limited to reported assays/doses and short follow-up.
  • AI role: The model generated seed cell-penetrating sequences, predicted quantitative nuclear-delivery activity from peptide fingerprints, and iteratively optimized sequences for higher predicted activity subject to length, charge, similarity and solubility constraints. The team synthesized and experimentally assayed selected model-designed peptides; this linked the AI design to measured enhanced PMO delivery and animal reporter results.

Evidence sources: https://pmc.ncbi.nlm.nih.gov/articles/PMC8819921/

Source checked · editorial details

The source page was checked against this entry. Independent confirmation is not assessed by this label.

Source type: paper. Paper: linked separately.

Review scope and date are not recorded; source checked does not establish full-text review.

Citations: in the top 10% of works of the same type, year and subfield (OpenAlex work metrics, 86 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.

Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.

Follow-up and relationships

No follow-up recorded; this is not a claim that no follow-up exists.

Follow-up review scope and date are not recorded.

No source-supported relationships recorded.

Open in the full catalog for filters, charts and related discoveries.