EBD-DTI is presented, a framework that enables zero-shot inference in graph-based DTI models without requiring any known interactions for unseen entities, with episodic cold-start training improving AUC by up to 12%.
Abstract
Predicting drug–target interactions (DTI) for entirely unseen drugs or proteins—the cold-start problem—remains a critical challenge in computational drug discovery. While sequence-based methods naturally support zero-shot generalization, they often ignore relational topology, and existing graph-based approaches either rely on global diffusion that blurs the boundary between inductive and transductive evaluation or require a few known interaction samples at test time (few-shot). We present EBD-DTI, a framework that enables zero-shot inference in graph-based DTI models without requiring any known interactions for unseen entities. The key innovation is episodic cold-start training : at each epoch, a random subset of training entities is masked and treated as pseudo-cold, forcing the model to learn cold-start inference with explicit gradient supervision. A bridge-conditioned local subgraph, together with multi-hop diffusion, provides cold entities with relational context from their nearest observed neighbors. Experiments on three benchmarks (BioSNAP, BindingDB, and DrugBank) demonstrate that EBD-DTI achieves competitive or superior performance compared to state-of-the-art methods under strict zero-shot evaluation, with episodic training improving AUC by up to 12%.
The proposed ZSCAN-DDIE framework enhances prediction accuracy for unseen DDIE categories but also provides biologically meaningful insights into molecular interaction mechanisms, offering a robust and clinically relevant solution for pharmacovigilance and drug safety assessment.
Shaoxi Gao, Zhanpeng Gan, Fangfang Han et al.· Bioinformatics Advances· 0 citations
The accurate identification of drug–drug interaction (DDI) events (DDIEs) is essential for ensuring medication safety and preventing adverse effects. However, novel drug development constantly generates new DDIE classes that suffer from a severe scarcity of labels, making zero-shot learning an essential solution for emerging cases. Unfortunately, the long-tailed distribution of existing DDI datasets causes overfitting to head classes and poor generalization to tail classes, thereby severely impairing zero-shot performance. To address this limitation, we propose a novel framework, ProAlign-DDI, which integrates a Substructure-Guided Semantic Adaptive Aggregation Module (SAAM) with a Prototype Alignment Module (PAM). In this framework, SAAM adaptively captures task-relevant features to produce high-quality aggregated representations. Subsequently, PAM employs class-level semantic prototypes to regularize these representations. This alignment strategy not only optimizes the global feature space but also alleviates the bias inherent in long-tailed DDI datasets. Comprehensive experiments show that ProAlign-DDI achieves competitive performance in conventional zero-shot learning and more balanced performance in generalized zero-shot learning, confirming its effectiveness in inferring unseen DDIE classes.
Unknown authors· Journal of Chemical Informat...· 0 citations
Computational drug–target interaction (DTI) prediction provides a scalable alternative to costly and time-consuming experimental screening, but its reliability is limited by the scarcity of experimentally verified negative interactions. In public DTI databases, most unobserved drug–target pairs are unlabeled rather than true non-interactions. Randomly treating these unlabeled pairs as negatives can introduce label noise and reduce model reliability, particularly in cold-start scenarios involving unseen drugs or targets. To address this issue, we propose LNMGAT, a LapRLS-guided reliable pseudo-negative mining framework coupled with dual graph attention encoders. Instead of relying on experimentally confirmed negative labels or randomly sampled negatives, LNMGAT first applies Laplacian regularized least squares to drug and target similarity graphs to identify low-confidence unlabeled pairs as reliable pseudo-negatives. Drug and target representations are then learned separately on similarity-based k-nearest-neighbor graphs using graph attention networks, and their embeddings are concatenated for MLP-based interaction prediction. Across Yamanishi, Davis, KIBA, and BindingDB benchmarks, LNMGAT achieved the best AUPR in 10 of 16 evaluation settings and ranked within the top two in 14 of 16 settings. In the 12 cold-start settings, LNMGAT obtained the best AUPR in 9 cases, with absolute AUPR gains over the strongest baseline of up to 0.016 in pair cold-start prediction. External evaluation on DrugBank positive interactions and SwissDock-based molecular docking further provided database-level and in silico support for the plausibility of high-ranked predictions. Nevertheless, the biological validation in this study remains computational and database-based; no wet-lab binding assay was performed. LNMGAT therefore provides a competitive and interpretable framework for DTI prediction under negative-label uncertainty, while further experimental validation is required for its top-ranked candidates.
Shuai Guo, Weichi Liu, Jie Zou et al.· Frontiers in Bioinformatics· 0 citations
Predicting drug synergy is essential for optimizing combination therapies in cancer treatment. Under extreme data scarcity, existing computational methods struggle to generalize to new cell lines. Although meta-learning approaches have shown promise, a critical limitation lies in their reliance on a unimodal Gaussian prior for task representation. In extreme few-shot settings, this assumption can overly pull the task posterior toward the prior mean, reducing the discriminability of task representations and pushing the model toward a generic mean-value predictor. To overcome this, we propose DiffPro, a novel framework that integrates a Latent Diffusion Model (LDM) as a structural prior. Unlike Gaussian-based methods, our diffusion module learns a flexible, data-driven prior that can model highly complex task distributions. This learned prior better reflects task heterogeneity in few-shot drug synergy prediction under extreme data scarcity, yielding more discriminative task representations. Comprehensive experiments on the DrugComb dataset demonstrate that DiffPro achieves a significant improvement in performance. Notably, in the challenging 5-shot setting, it outperforms the best-performing baseline, delivering approximately a 17.3% relative gain in R2 (0.264 vs. 0.225) and a 4.6% relative reduction in MSE. These results confirm that the diffusion prior successfully regularizes the latent space, mitigating task representation collapse and enabling robust, high-precision predictions for combination therapy design.
Shuting Jin, Xu Guo, Anqi Huang et al.· Proceedings of the 32nd ACM...· 0 citations
DQHTFI is proposed, a fine-grained interaction prediction framework for drug–target interaction classification and binding affinity regression that employs BRICS fragments and Pfam functional domains as the basic interaction units and jointly learns semantic and structural representations.
Ke Tao, Haoze Chai, Zhuo Chen et al.· bioRxiv· 0 citations
Drug–target interaction (DTI) prediction is a critical step in drug discovery, and accurate prediction of potential interactions can significantly accelerate the drug-development process. Although deep-learning approaches have achieved promising performance in DTI prediction, two challenges remain: single models often fail to comprehensively capture heterogeneous sequence information, resulting in limited stability and generalization, while insufficient integration of local and global features restricts interaction representation. To address these limitations, we propose HFEDTI, a DTI prediction model that integrates hierarchical feature fusion and weighted ensemble learning. Specifically, a residual convolutional neural network (ResCNN) is employed to extract local structural features of drugs and targets, while a self-attention-based hierarchical bidirectional long short-term memory network (SAHBiLSTM) captures global contextual dependencies. Furthermore, a hierarchical heterogeneous attention mechanism is introduced to align and fuse multi-level cross-modal representations, and a weighted ensemble strategy based on validation performance ranking is developed to enhance model robustness and generalization. Experimental results on three benchmark datasets demonstrate the effectiveness of HFEDTI. On the DrugBank dataset, HFEDTI achieves an AUC of 0.9238 and an AUPR of 0.9327, improving the best-performing baseline by 0.90 and 1.40 percentage points, respectively. Moreover, HFEDTI consistently achieves strong performance on the C. elegans and Human datasets, further validating its effectiveness and generalization capability for DTI prediction.
Unknown authors· Electronics· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
