Abstract:Abstract A major concern with co-administration of different drugs is the high risk of interference between their mechanisms of action, known as adverse drug–drug interactions (DDIs), which can cause serious injuries to the organism. Although several computational methods have been proposed for identifying potential adverse DDIs, there is still room for improvement. Existing methods are not explicitly based on the knowledge that DDIs are fundamentally caused by chemical substructure interactions instead of whole drugs’ chemical structures. Furthermore, most of existing methods rely on manually engineered molecular representation, which is limited by the domain expert’s knowledge.We propose substructure–substructure interaction–drug–drug interaction (SSI–DDI), a deep learning framework, which operates directly on the raw molecular graph representations of drugs for richer feature extraction; and, most importantly, breaks the DDI prediction task between two drugs down to identifying pairwise interactions between their respective substructures. SSI–DDI is evaluated on real-world data and improves DDI prediction performance compared to state-of-the-art methods. Source code is freely available at https://github.com/kanz76/SSI-DDI.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the prediction of drug - drug interactions (DDIs). Specifically, the paper points out that when different drugs are used simultaneously, adverse interactions may occur between them, and these interactions may lead to serious health risks. Although a variety of computational methods have been proposed to identify potential adverse DDIs, there is still room for improvement in existing methods. The main deficiencies are as follows: 1. **Existing methods are not based on the root causes of DDIs**: that is, the interactions between chemical sub - structures, rather than the chemical structures of the entire drugs. 2. **Dependence on artificially - designed molecular representations**: This limits the effectiveness of the method because such representations are limited by the knowledge of domain experts and may contain information that is useless for downstream tasks. To solve these problems, the paper proposes a new deep - learning framework - Substructure - Substructure Interaction - Drug - Drug Interaction (SSI - DDI). This framework directly operates on the original molecular graph representations of drugs to extract richer features and decomposes the DDI prediction task into identifying pairwise interactions of sub - structures between two drugs. Through this method, SSI - DDI outperforms the existing state - of - the - art methods on real - world datasets. ### Main contributions of the paper: 1. **Knowledge - driven model**: Explicitly treat sub - structures as independent entities and determine whether a drug may have a potential DDI through their pairwise interactions. 2. **Deep - learning model**: Directly learn sub - structures from the molecular graphs of drugs and use the co - attention mechanism to learn the relevance or importance of each sub - structure interaction to the final DDI prediction. This not only improves the prediction accuracy but also enhances the interpretability of the results. ### Method overview: - **Input representation**: Drugs are represented as a graph \(G=(V, E)\), where \(V\) is the set of nodes and \(E\) is the set of edges. Each interaction type \(I_r\) is represented as a learnable matrix \(M_r\). - **Graph convolution operator**: Use the Graph Attention Network (GAT) layer for node feature update and stabilize the learning process through the multi - head attention mechanism. - **Sub - structure extraction**: Aggregate information of different sub - structures through convolution operations, and each sub - structure consists of a node and its neighboring nodes. - **Relevance of sub - structure interactions**: Use the co - attention mechanism to calculate the importance of the interactions between each sub - structure pair. - **Prediction and loss function**: Predict the probability of DDI by aggregating weighted sub - structure interaction scores and use the binary cross - entropy loss function for end - to - end training. ### Experimental results: - **Dataset**: Use the DrugBank dataset (version 5.0.3), which contains 1,704 drugs and 191,400 DDIs. - **Performance evaluation**: Evaluated on multiple metrics (accuracy ACC, area under the receiver operating characteristic curve AUROC, area under the precision - recall curve AUPRC), and the results show that SSI - DDI significantly outperforms other methods on ACC and AUPRC. - **Inductive setting**: In the cold - start scenario, that is, when predicting unseen drugs, SSI - DDI still shows good performance, especially on the AUPRC metric. In conclusion, by introducing the concept of sub - structure - sub - structure interaction, this paper provides a more effective and interpretable DDI prediction method, which is expected to play an important role in drug development and clinical applications.

SSI–DDI: substructure–substructure interactions for drug–drug interaction prediction

SSF-DDI: a deep learning method utilizing drug sequence and substructure features for drug–drug interaction prediction

STNN-DDI: A Substructure-aware Tensor Neural Network to Predict Drug-Drug Interactions

Molecular Substructure-Aware Network for Drug-Drug Interaction Prediction

Multi-View Substructure Learning for Drug-Drug Interaction Prediction

An effective framework for predicting drug–drug interactions based on molecular substructures and knowledge graph neural network

Learning size-adaptive molecular substructures for explainable drug–drug interaction prediction by substructure-aware graph neural network

DSN-DDI: an accurate and generalized framework for drug–drug interaction prediction by dual-view representation learning

Drug–drug interaction prediction with learnable size-adaptive molecular substructures

Molecular Structure-Based Double-Central Drug-Drug Interaction Prediction

MSDAFL: molecular substructure-based dual attention feature learning framework for predicting drug-drug interactions

DSIL-DDI: A Domain-Invariant Substructure Interaction Learning for Generalizable Drug–Drug Interaction Prediction

A Molecular Fragment Representation Learning Framework for Drug–Drug Interaction Prediction

ISCMF: Integrated similarity-constrained matrix factorization for drug–drug interaction prediction

MASMDDI: multi-layer adaptive soft-mask graph neural network for drug-drug interaction prediction

Learning self-supervised molecular representations for drug–drug interaction prediction

CASTER: Predicting Drug Interactions with Chemical Substructure Representation

AttentionDDI: Siamese attention-based deep learning method for drug–drug interaction predictions

Deep learning for drug‐drug interaction prediction: A comprehensive review

StructNet-DDI: Molecular Structure Characterization-Based ResNet for Prediction of Drug–Drug Interactions

Multitype drug interaction prediction based on the deep fusion of drug features and topological relationships