Feedback-MitAptamerGAN for SELEX-informed generation and prioritization of mitochondrial RNA aptamer candidates
Background Designing mitochondrial RNA aptamers requires a way to translate experimentally enriched sequence patterns into new candidates with measurable binding. Methods We developed Feedback-MitAptamerGAN, which couples a Wasserstein generative adversarial network with an evaluator combining ERNIE-RNA representations and RNAfold features. The model draws on three published aptamers and an initial resource of 20,000 unique sequences recovered after three rounds of selection against HEK293-derived mitochondria. A separate evaluator uses continuous source-informed targets, including later-round SELEX sequences, to guide candidate generation. Results Across three matched experimental batches, the complete-model library maintained the lowest mean cycle threshold in every round and reached 14.37 ± 0.20 at round 6, compared with 16.93–18.45 for the ablated libraries and 28.83 ± 0.17 for the random library. Values are means ± sample SD across independently prepared selection runs, each measured in technical triplicate. At a BLI-selected checkpoint, ten high-score candidates had a mean apparent dissociation constant of 59.06 nominal nM, compared with 8,330.37 nominal nM for ten lower-score candidates, based on three independent experiments per sequence. MitApt-A02 pull-down material contained VDAC1. Conclusion Together, these results support feedback-guided generation and prioritization of mitochondrial RNA aptamer candidates.
Authors
- Zidong Ji
- Yueling Long
- Zhongyun Chen
Institutions
- Hunan University (CN)
Publication Details
- Journal
- Frontiers in Bioengineering and Biotechnology
- Published
- 2026-11-16
- DOI
- https://doi.org/10.3389/fbioe.2026.2016105
- Primary Topic
- RNA and protein synthesis mechanisms
- Type
- article
- Field-Weighted Citation Impact
- 0.00
Funders
- Natural Science Foundation of Hunan Province