EUVIMEDEUVIMED entdecken
Open Access

Sequence-Driven Classification of Aptamers by Target Type: Algorithm Development and Validation

Ava George, Andrew Takayama, Andres Medina, Amanda Reyes, Damian Ventura, Stevan Pecic, Ankita Mohapatra

JMIR Bioinformatics and Biotechnology · 13.11.2026 · 7 · e92194-e92194

DOI: 10.2196/92194 ↗

Kurz gefasst

Fachartikel in der Zeitschrift JMIR Bioinformatics and Biotechnology, online erschienen am 13.11.2026. Verfasst von Ava George und 6 weiteren Autor:innen. Beteiligte Einrichtung: Department of Computer Science, Fullerton College, Fullerton, CA, United States. Die Arbeit nennt 20 Literaturangaben. Der Volltext ist frei in PubMed Central verfügbar.

AbstractSchlagwörterBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Background

DNA aptamers are short, single-stranded oligonucleotides that bind molecular targets with high affinity and specificity. Traditionally, aptamers are identified through SELEX (systematic evolution of ligands by exponential enrichment), an iterative process that is time-consuming and requires specialized equipment. To simplify and accelerate aptamer discovery, particularly across multiple molecular targets, computational approaches have gained momentum.

Objective

In this study, we present a sequence-only framework for classifying DNA aptamers by target type using curated datasets of experimentally validated sequences. Target type here refers to the broad class of ligands that an aptamer binds to.

Methods

Aptamer-target records were curated from the UTexas Aptamer Database, retaining only those sequences with complete annotations and a clearly defined target category. From these filtered records, the 2 target categories with the highest number of unique aptamer sequences—cell and steroid, comprising a total of 236 sequences—were selected for this study. Each sequence was then represented using character-level term frequency–inverse document frequency of n-gram motifs (n=3‐6) combined with guanine-cytosine content and sequence length. The dataset was then split into 70% training, 20% testing, and 10% validation sets. An ensemble of 9 bagged extreme gradient boosting classifiers was selected for its established performance on high-dimensional tabular data and native handling of class imbalance. The model was trained to distinguish between aptamers targeting cellular and steroidal molecules, with class weighting, oversampling, probability calibration, and per-class threshold optimization applied to address class imbalance. Model performance was evaluated using accuracy, macro- F 1 scores, and recall on the test dataset and an independently generated external validation set comprising 13 new laboratory-generated steroid (aldosterone)–binding aptamers. The novelty of these laboratory aptamers relative to the training set was also confirmed using the pairwise Levenshtein distance.

Results

The ensemble classifier achieved 97.92% accuracy and a macro- F 1 -score of 0.95 on a test set of 236 curated DNA aptamer sequences, demonstrating strong performance in distinguishing cell-binding from steroid-binding aptamers. The calibrated and threshold-optimized model was further validated on a set of 13 new laboratory-created aptamers for steroid aldosterone using experimental SELEX, achieving 100% recall and confirming its ability to generalize to new experimental data.

Conclusions

These results demonstrate that informative, target-specific patterns can be directly extracted from DNA sequences, establishing a scalable, data-driven complement to SELEX in this preliminary proof-of-concept study for rapid prescreening of candidate aptamers. More broadly, this sequence-only approach could support faster, lower-cost triage of candidate aptamers, reducing reliance on iterative wet-laboratory SELEX as aptamer databases and target categories continue to expand.

Schlagwörter

Stichwörter der Autor:innen

DNA aptamersSELEXXGBoostdecision treesextreme gradient boostinginverse document frequencymachine learningsequence-based classificationsystematic evolution of ligands by exponential enrichmentterm frequency

Autor:innen und Institutionen

  • Department of Computer Science, Fullerton College, Fullerton, CA, United States.
  • Department of Chemistry and Biochemistry, California State University, Fullerton, CA, United States.
  • Department of Electrical and Computer Engineering, California State University, 500 N State Clg Blvd, Fullerton, CA, 92831, United States, +1 (657) 278-2853.

Zitieren

APA

George, A., Takayama, A., Medina, A., Reyes, A., Ventura, D., Pecic, S., & Mohapatra, A. (2026). Sequence-Driven Classification of Aptamers by Target Type: Algorithm Development and Validation. JMIR Bioinformatics and Biotechnology, 7, e92194-e92194. https://doi.org/10.2196/92194

Vancouver

George A, Takayama A, Medina A, Reyes A, Ventura D, Pecic S, et al. Sequence-Driven Classification of Aptamers by Target Type: Algorithm Development and Validation. JMIR Bioinformatics and Biotechnology. 2026;7:e92194-e92194. doi:10.2196/92194

BibTeX
@article{George2026,
  author = {George, Ava and Takayama, Andrew and Medina, Andres and Reyes, Amanda and Ventura, Damian and Pecic, Stevan and Mohapatra, Ankita},
  title = {Sequence-Driven Classification of Aptamers by Target Type: Algorithm Development and Validation},
  journal = {JMIR Bioinformatics and Biotechnology},
  year = {2026},
  volume = {7},
  pages = {e92194--e92194},
  doi = {10.2196/92194}
}
Kostenlos starten →