Use OligoAI Online
Commercially Available OligoAI No-Code Web Server
OligoAI: Predicting Antisense Oligonucleotide Knockdown Efficacy with Deep Learning
OligoAI is a transformer-based deep learning model that predicts the knockdown efficacy of RNase H-dependent antisense oligonucleotide (ASO) gapmers from the ASO sequence, its surrounding target RNA context, chemical modification pattern, dosage and transfection method. It was introduced in a 2025 bioRxiv preprint alongside ASO Atlas, a large patent-derived dataset of ASO screens. The authors host a web tool for OligoAI at sitlabs.org/OligoAI and state that they are making the database and tool freely available; the pages we could read did not give author names, affiliations or a license.
Choosing an effective ASO means choosing both the sequence and the chemical modification pattern, and experimental screening is costly. Computational tools have been limited by a lack of large, standardized training data, so earlier predictors generalize poorly across targets and chemistries.
Attribute | Detail |
|---|---|
Task | Predict in vitro knockdown of RNase H gapmer ASOs |
Sequence encoder | Pre-trained RiNALMo-giga on the ASO plus 50 nt of target context on each side |
Other inputs | Learned chemistry and backbone embeddings, dosage, transfection method |
Training data | ASO Atlas: 188,521 gapmers from 417 patents, 334 gene targets |
Held-out performance | Median Spearman 0.419 across 299 screens, 3.14x enrichment |
How OligoAI Models ASO Sequence, Chemistry and Context
The model combines a language-model view of the RNA with explicit experimental conditions, then regresses to a knockdown value.
Encode sequence and context: RiNALMo-giga embeds the ASO and the target region surrounding it.
Add chemistry and conditions: learned embeddings for chemistry and backbone are joined with dosage and transfection method.
Predict knockdown: a multilayer perceptron outputs the predicted knockdown.
Building the ASO Atlas
The dataset was assembled by extracting screening tables from USPTO patents (2001 to 2025) with LLM assistance, restricted to ISIS or IONIS Pharmaceuticals filings. Among ASOs with chemistry annotations, 52.6% were 5-10-5 2'-MOE and 30.4% were 3-10-3 cEt. Median knockdown was 45.0%, and 17.4% of ASOs showed under 10% knockdown. The authors also report that splice-junction ASOs had the lowest median knockdown (32.0%), 3'UTR ASOs the highest (53.0%), unpaired targets showed 8.5% higher knockdown than base-paired ones, and the GGGG motif was linked to lower efficacy.
Benchmarks and Experimental Validation of OligoAI
Evaluation | Result |
|---|---|
Held-out screens (299) | OligoAI Spearman 0.419; ASOptimizer 0.076; OligoWalk 0.147 |
PSD3 holdout (N = 2,234) | Top 1% ranked ASOs: median knockdown 92.5% vs 65.0% overall |
KCNT2 in HeLa cells | Top-ranked ASOs 81% median knockdown vs 36% for random (P = 1.1e-4) |
Screening effort | Estimated 5.72-fold reduction (95% CI 2.94 to 9.94) |
For the KCNT2 test, a virtual library of 200,374 20-mer gapmers was screened, and 18 top-1% OligoAI picks and 32 random ASOs were tested experimentally.
What is Tamarind Bio?
Tamarind Bio is a no-code bioinformatics platform built to give life scientists and researchers access to powerful computational tools. Many cutting-edge machine learning models are hard to deploy and use. Tamarind provides an intuitive, web-based environment that removes the complexity of high-performance computing, software dependencies and command-line interfaces.
The platform is designed for biologists, chemists and other researchers who may not have a background in programming or cloud infrastructure but want to run models on their own data. Key features include:
A user-friendly graphical interface for setting up and launching experiments
A robust API for integration into existing research pipelines
An automated system for managing and scaling computational resources
Tamarind treats information and data security as a top priority, as detailed in its Trust Center and Terms of Service.
Accelerating Discovery with OligoAI on Tamarind Bio
Virtual ASO screening: rank large libraries of candidate gapmers for a target and send only the top fraction to the bench.
Chemistry comparison: compare predicted knockdown across modification patterns such as 5-10-5 2'-MOE and 3-10-3 cEt.
Target-site selection: use target context to avoid low-efficacy regions and motifs.
Lower screening cost: the authors estimate a 5.72-fold reduction in screening effort.
How to Use OligoAI on Tamarind Bio
Open the tool: log in to tamarind.bio and open OligoAI.
Provide the target: supply the target RNA sequence so the model can use the surrounding context.
Define candidates: enter the ASO sequences or generate a library of gapmers across the target.
Set conditions: if exposed, choose chemistry and backbone, dosage and transfection method.
Run the prediction: submit the job.
Download outputs: save the predicted knockdown for each ASO.
Prioritize: rank candidates, select the top fraction and validate experimentally.
Things to Keep in Mind
Only RNase H-dependent gapmers are covered.
Training data comes from a single company (IONIS/ISIS) and may be biased.
LLM-based annotation introduced some errors; the paper cites a chemistry error rate of about 3 to 4%.
Predictions are in vitro only; in vivo efficacy, off-target effects and toxicity are not modeled.
Experimental validation covered one gene (KCNT2) in one cell line (HeLa).
Source: OligoAI and ASO Atlas preprint, bioRxiv, 2025, DOI 10.1101/2025.10.29.685292, https://www.biorxiv.org/content/10.1101/2025.10.29.685292v1. Web tool: https://sitlabs.org/OligoAI.