Abstract:Motivation Small interfering RNA (siRNA) is often used for function study and expression regulation of specific genes, as well as the development of small molecule drugs. Selecting siRNAs with high inhibition and low off-target effects from massive candidates is always a great challenge. Increasing experimentally validated samples prompt the development of machine-learning-based algorithms, including Support Vector Machine (SVM), Convolutional Neural Network (CNN), and Graph Neural Network (GNN). However, these methods still suffer from limited accuracy and poor generalization to design both potent and specific siRNAs. Results In this study, we propose a novel approach for siRNA inhibition and off-target effect prediction, named AttSiOff. It combines self-attention-based siRNA inhibition predictor with an mRNA searching package and an off-target filter. The predictor gives the inhibition score via analyzing the embedding of siRNA and local mRNA sequences, generated from pre-trained RNA-FM model, as well as other meaningful prior-knowledge-based features. Self-attention mechanism can detect potentially decisive features, which may determine the inhibition of siRNA. It captures global and local dependencies more efficiently than normal convolutions. The 10-fold cross-validation results indicate that our model achieves a significant improvement of correlation between prediction and label, compared with all existing methods. And it reaches better performance of generalization and robustness on cross-dataset validation. In addition, the mRNA searching package could find all mature mRNAs for given gene name from GENOMES database, and the off-target filter can calculate the amount of unwanted off-target binding sites, which affects the specificity of siRNA. Experiments on mature siRNA drugs show that our entire framework, AttSioff, have excellent convenience and operability in practical applications. Contact yuanye_auto{at}sjtu.edu.cn or chengjin520{at}sjtu.edu.cn. ### Competing Interest Statement The authors have declared no competing interest. * RNA-FM : RNA fundamental model RNAi : RNA interfering siRNA : small interfering RNA CNN : Convolutional Neural Network GNN : Graph Neural Network PCC : Pearson Correlation Coefficient SPCC : Spearman Correlation Coefficient AUC : the Area under the Receiver Operating Characteristic curve

AttSiOff: A self-attention-based approach on siRNA design with inhibition and off-target effect prediction

siRNADesign: A Graph Neural Network for siRNA Efficacy Prediction via Deep RNA Sequence Analysis

Effective small interfering RNA design based on convolutional neural network

siRNADiscovery: a graph neural network for siRNA efficacy prediction via deep RNA sequence analysis

OligoFormer: an accurate and robust prediction method for siRNA design

SiRNA Silencing Efficacy Prediction Based on a Deep Architecture

Influence of mRNA features on siRNA interference efficacy.

siRNA online design system based on multi-machine learning methods

A Computational Model for Compressed Sensing RNAi Cellular Screening.

Batch RNAi Selector: A Standalone Program to Predict Specific Sirna Candidates in Batches with Enhanced Sensitivity.

In Silico Target-Specific Sirna Design Based on Domain Transfer in Heterogeneous Data.

Optimized sgRNA design by deep learning to balance the off-target effects and on-target activity of CRISPR/Cas9

Prediction of siRNA Efficacy Using BP Neural Network and Support Vector Machine

Multi-task learning for cross-platform siRNA efficacy prediction: an in-silico study

Mitigating off‐target effects of small RNAs: conventional approaches, network theory and artificial intelligence

Prediction of Off-Target Specificity and Cell-Specific Fitness of CRISPR-Cas System Using Attention Boosted Deep Learning and Network-Based Gene Feature.

Reconsideration of In-Silico Sirna Design Based on Feature Selection: A Cross-Platform Data Integration Perspective

Support Vector Machine Approach For Sirna Functionality Prediction

A semi–supervised tensor regression model for siRNA efficacy prediction

BERT-siRNA: siRNA target prediction based on BERT pre-trained interpretable model

Support Vector Machine for Prediction of Sirna Silencing Efficacy