Structure-informed models for ionic current prediction in nanopore sequencing of expanded dna alphabets

Area of Science:

• Biotechnology
• Bioinformatics
• Synthetic Biology

Background:

• Nanopore sequencing offers direct single-molecule analysis of biopolymers like DNA and RNA.
• Xenonucleic acids (XNAs) expand DNA capabilities for diagnostics, therapeutics, and data storage.
• Current nanopore sequencing methods for modified bases require extensive, costly experimental calibration.

Purpose of the Study:

• To develop computational methods for predicting ionic current signals during nanopore sequencing of XNA.
• To reduce the reliance on empirical calibration for sequencing noncanonical DNA bases.
• To assess the effectiveness of sequence-based versus structure-aware predictive models.

Main Methods:

• Compared a sequence-based predictive model against two structure-aware models.
• Utilized graph-based molecular representations for one structure-aware approach.
• Adapted a generative language model using molecular SMILES for another structure-aware approach.

Main Results:

• Sequence context explains significant signal variability in XNA nanopore sequencing.
• Structure-aware models, incorporating chemical and structural information, improved predictive accuracy.
• The findings demonstrate the potential for computational approaches to enhance XNA sequencing scalability.

Conclusions:

• Computational modeling can significantly reduce experimental calibration for XNA nanopore sequencing.
• Integrating structural and chemical data into predictive models is crucial for accuracy.
• This framework may be applicable to modeling ionic currents for other complex biomolecules, including proteins.