Regression-Based Modeling of Antisense Oligonucleotide Efficacy Using Sequence, Structural, and Off-Target Features

Ava George, Yu Bai, Jialin Tang

Abstract:

Antisense oligonucleotides (ASOs) are a
promising class of nucleic acid–based therapeutics
that regulate gene expression by binding target
mRNAs, with applications in genetic and rare
diseases. However, designing effective ASOs remains
difficult due to the vast combinatorial space of
sequences, secondary structures, and chemical
modifications. Recent work has leveraged deep
learning and graph neural networks to address these
challenges. Building on this foundation, the present
project explores a complementary pipeline using
classical machine learning and statistical methods for
ASO design and evaluation. The workflow integrates
multiple computational stages: retrieval of target
mRNA sequences from NCBI, interaction prediction
using the miRanda algorithm, and structural analysis
via ViennaRNA. Off-target interactions were
systematically assessed, and custom Python scripts
were developed to merge outputs into a unified
dataset. Feature engineering incorporated both
numeric and categorical predictors, such as cell line
and density, enabling model testing of inhibitory
efficiency. Features of sequence, structure and offtarget interactions trained multiple regressors
including Linear Regression, Ridge, Lasso, Random
Forest, and Gradient Boosting. Models were
evaluated using nested cross-validation with groupaware splits to prevent leakage. Random Forest
achieved the highest predictive performance to
predict inhibition outcomes (R² ≈ 0.627, MAE ≈
9.47). These results highlight both the feasibility and
the challenges of applying interpretable machine
learning techniques to ASO design, particularly in the
presence of substantial missing data. Future
directions include exploring a normalized
hybridization energy gradient with relative energy
per nucleotide. This work demonstrates the potential
for combining bioinformatics tools, structural
modeling, and machine learning to advance the
rational design of therapeutic ASOs.