ASODrugDesignAI: AI-Accelerated Oligonucleotide Drug Design

Machine learning (ML) and Artificial Intelligence (AI) have been a remarkable success in improving small molecule drug discovery over the past decade. Advanced ML models have been applied to generate novel molecular structures, suggest synthesis pathways, or predict biochemical properties or interactions of those molecules. While small molecules interacting with proteins are the most frequently used drug modality today, alternatives have been identified to address unmet clinical needs. In particular advances in chemistry for drugs based on oligonucleotides have shown promise to deliver successful therapies to the clinic. Common to oligonucleotide drugs is that they target the expression of genes, for example by leveraging inter-cellular mechanisms to degrade the gene's transcript, i.e., its mRNA, before it is translated to a protein. The fundamental reaction is hybridization of a single-stranded oligonucleotide molecule (most commonly DNA) to either single-stranded RNA or single/double-stranded DNA. The main determinant of the reaction, important for drug efficacy and drug safety, are the thermodynamics of oligonucleotide binding which has been very well studied for the case of DNA-DNA binding with ample data available.

The clinical use of oligonucleotides requires chemical modifications to increase stability and reduce toxicity; out of the vast spectrum of possible modifications only a very limited amount has been experimentally investigated with respect to impact on thermodynamics. Existing data suggests that alternative modifications can have a sizable effect on safety of drugs. Additionally, the efficiency with which oligonucleotides can pass the cell membrane and be transported to the appropriate location in the cell can be increased with conjugation (binding) to specifically designed small molecules. Also here, prediction of thermodynamics of conjugated oligonucleotides is needed.

The focus on project is on the following aims.

Predicting thermodynamic effects of novel modifications to oligonucleotides. Driven by the ample data for thermodynamics of DNA-DNA hybridization, we will develop ML models which can serve as the basis for transfer to modified oligonucleotides. Modifications such as linked nucleic acids (LNA), for which data is available due to utilization in drugs under development, will provide the data for transfer learning. Extensions to novel modifications will be additionally based on in silico simulations for modifications where experimental data is sparse

Predicting thermodynamic effects of conjugation on thermodynamic of oligonucleotide binding We will collect existing data on conjugated oligonucleotides and information on their transport and their binding affinity or thermodynamics. Additional sources of information will be RNASeq data for efficacy and safety of conjugated oligonucleotides and chemistry simulations. The goal is to predict suitable conjugants dependent on oligonucleotide sequence and targeted cell type.

Enable federated, privacy-preserving learning of thermodynamics prediction. Federated learning, already in use in industry for small molecule drug design, allows competing entities to learn ML models for tasks such as off-target binding prediction from pooled data without comprising privacy of drug candidate data.

The project is funded by WASP-DDLS

For further information contact Alexander Schliep (


Members: Alexander Schliep, Shirin Tavara. Collaborators: Pär Mattson (University of Gothenburg), Rula Zain (Karolinska Institutet), Antonio Carlesso (University of Gothenburg).


Tavara et al.. Federated Learning of Oligonucleotide Drug Molecule Thermodynamics with Differentially Private ADMM-Based SVM. In Machine Learning and Principles and Practice of Knowledge Discovery in Databases: International Workshops of ECML PKDD 2021, Virtual Event, September 13-17, 2021, Proceedings, Part II, Springer, 459–467, 2022.