Organizations: School of Computer Science and Engineering, Northeastern University, Shenyang 110819, China · College of Information Science and Engineering, Northeastern University, Shenyang 110819, China · Westlake University, Hangzhou 310000, China · School of Artificial Intelligence, Beihang University, Beijing 100000, China · Key Laboratory of Bioresource Research and Development of Liaoning Province, College of Life and Health Sciences, Northeastern University, Shenyang 110169, China
Predicting drug-target affinity is fundamental to virtual screening and lead optimization. However, existing deep models often suffer from representation collapse in stringent cold-start regimes, where the scarcity of labels and domain shifts prevent the learning of transferable pharmacophores and binding motifs. In this paper, we propose Co-Diffusion, a novel affinity-aware framework that redefines DTA prediction as a constrained latent denoising process to enhance generalization. Co-Diffusion employs a two-stage paradigm: Stage I establishes an affinity-steered latent manifold by aligning drug and target embeddings under an explicit supervised objective, ensuring that the latent space reflects the intrinsic binding landscape. Stage II introduces modality-specific latent diffusion as a stochastic perturb-and-denoise regularizer, forcing the model to recover consistent affinity semantics from noisy structural representations. This approach effectively mitigates the reconstruction-regression conflict common in generative DTA models. Theoretically, we show that Co-Diffusion maximizes a variational lower bound on the joint likelihood of drug structures, protein sequences, and binding strength. Extensive experiments across multiple benchmarks demonstrate that Co-Diffusion significantly outperforms state-of-the-art baselines, particularly yielding superior zero-shot generalization on unseen molecular scaffolds and novel protein families-paving a robust path for in silico drug prioritization in unexplored chemical spaces.
Figures & tables
Fig. 1 : Overview of the drug discovery process. The proposed Co-Diffusion framework operates at the predictive modeling stage, providing robust affinity estimates to guide the selection of candidates before in vitro testing.
Fig. 2 : Latent Diffusion Process in Training and Inference. (a) In training, the VAE encoder produces latent z0 , which is perturbed to zt by adding noise. A UNet is trained to predict the noise, following the Noise-Prediction Training Objective . (b) In inference, the reverse denoising process transforms sampled noise zT back to z0 , which is decoded to generate the target sequence, as described in Generation .
Fig. 3 : The graphical structure for Co-Diffusion model.
Fig. 4 : Overview of Co-Diffusion.
Dataset
Targets
Drugs
Binding pairs
KIBA
229
2111
118,254
Davis
442
68
30,056
TABLE I : Statistics of the benchmark datasets used in this study.
Fig. 5 : Performance comparison on Davis and KIBA under random split.
Dataset
Settings
Method
MSE(std) ↓
MAE(std) ↓
CI(std) ↑
rm2↑
Davis
Unseen-drug
DeepDTA
0.745 (0.008)
0.704 (0.050)
0.699 (0.045)
0.078
AttentionDTA
0.656 (0.004)
0.622 (0.003)
0.682 (0.010)
0.104
GraphDTA
0.590 (0.016)
0.552 (0.014)
0.695 (0.008)
0.077
Co-VAE
0.577 (0.009)
0.523 (0.014)
0.616 (0.005)
0.128
TransVAE-DTA
0.643 (0.012)
0.542 (0.009)
0.543 (0.010)
0.007
Pair-VAE
0.570 (0.004)
0.498 (0.007)
0.730 (0.003)
0.134
TABLE II: Performance Comparison on Davis and KIBA Datasets Under Cold-start Scenarios.
Fig. 6 : Scatter plots of predicted versus actual affinity on Davis (top) and KIBA (bottom) under Unseen Drug, Unseen Target, and Unseen Pair. The dashed line denotes the identity line.
Settings
Method
MSE ↓
MAE ↓
CI(std) ↑
rm2↑
Unseen Drug
Co-Diffusion
0.568
0.471
0.722
0.166
Drug-Diffusion
0.580
0.498
0.630
0.122
Target-Diffusion
0.617
0.537
0.636
0.122
No-diffusion
0.571
0.519
0.619
0.131
Unseen Target
Co-Diffusion
0.556
0.429
0.836
0.461
Drug-Diffusion
0.542
0.482
0.821
0.457
TABLE III : Ablation test of diffusion components under cold-start scenarios on the Davis dataset.
Setting
Method (Co-Diffusion)
KIBA
Davis
MSE ↓
MAE ↓
CI ↑
RM2 ↑
MSE ↓
MAE ↓
CI ↑
rm2↑
Unseen Drug
end-to-end
0.537
0.490
0.686
0.256
0.622
0.509
0.692
0.127
two-stage
0.421
0.406
0.751
0.402
0.568
0.471
0.722
0.166
Unseen Target
end-to-end
0.405
0.439
0.632
0.290
0.522
0.413
0.830
0.496
two-stage
0.347
0.383
0.685
0.377
0.556
0.429
0.836
0.461
Unseen Pair
end-to-end
0.488
0.505
0.597
0.135
0.858
0.570
0.693
0.121
TABLE IV : Comparison of Co-Diffusion trained with an affinity-aware two-stage regimen versus an end-to-end variant across cold-start scenarios on KIBA and Davis.
T
Setting
Davis
KIBA
MAE ↓
rm2↑
MAE ↓
rm2↑
800
Unseen Pair
0.5487
0.1401
0.4669
0.2074
Unseen Drug
0.4709
0.1614
0.4096
0.4011
Unseen Target
0.4311
0.4477
0.3953
0.3801
1000
Unseen Pair
0.5510
0.1491
0.4591
0.2011
Unseen Drug
0.4710
0.1660
0.4065
0.4016
TABLE V : Sensitivity of Co-Diffusion to the number of diffusion steps T on Davis and KIBA.
Beta
Setting
Davis
KIBA
MAE ↓
rm2↑
MAE ↓
rm2↑
[1e−5,4e−5]
Unseen Pair
0.5842
0.1408
0.4713
0.1986
Unseen Drug
0.5100
0.1612
0.4062
0.3967
Unseen Target
0.4845
0.4317
0.4048
0.3557
[1e−4,4e−4]
Unseen Pair
0.5510
0.1491
0.4591
0.2011
Unseen Drug
0.4710
0.1660
0.4065
0.4016
TABLE VI : Sensitivity of Co-Diffusion to the number of noise magnitude beta on Davis and KIBA.
Fig. 7 : t-SNE projections of new (blue) vs. original (green) drug and target embeddings on KIBA (a–f) and Davis (g–l) under unseen drug, unseen target, and unseen pair scenarios.
ID
Year
True
Pred.
MSE ( ↓ )
6j0g
2019
5.640
5.609
0.001
6ce6
2018
5.100
5.162
0.004
5zyg
2019
5.210
5.277
0.005
6qlp
2019
5.280
5.349
0.005
6s56
2019
5.100
5.182
0.007
5xpi
2018
5.280
5.189
0.008
TABLE VII : Out-of-sample results on recent PDBbind entries.
Drug--target affinity prediction is pivotal for accelerating drug discovery, yet existing methods suffer from significant performance degradation in realistic cold-start scenarios (unseen drugs/targets/pairs), primarily driven by overfitting to training instances and information loss from irrelevant target sequences. In this paper, we propose LaPro-DTA, a framework designed to achieve robust and generalizable DTA prediction. To tackle overfitting, we devise a latent dual-view drug representation mechanism. It synergizes an instance-level view to capture fine-grained substructures with stochastic perturbation and a distribution-level view to distill generalized chemical scaffolds via semantic remapping, thereby enforcing the model to learn transferable structural rules rather than memorizing specific samples. To mitigate information loss, we introduce a salient protein feature extraction strategy using pattern-aware top-k pooling, which effectively filters background noise and isolates high-response bioactive regions. Furthermore, a cross-view multi-head attention mechanism fuses these purified features to model comprehensive interactions. Extensive experiments on benchmark datasets demonstrate that LaPro-DTA significantly outperforms state-of-the-art methods, achieving an 8% MSE reduction on the Davis dataset in the challenging unseen-drug setting, while offering interpretable insights into binding mechanisms.
Zihan Dun, Liuyi Xu, An-Yang Lu +2
College of Information Science and Engineering, Northeastern University, Shenyang 110819, China · School of Artificial Intelligence, Beihang University, Beijing 100000, China · School of Computer Science and Engineering, Northeastern University, Shenyang 110819, China
Predicting drug--target binding affinity (DTA) requires models to distinguish subtle chemical and structural determinants underlying molecular recognition. Although recent approaches increasingly incorporate richer drug and protein information, such information may be compressed, homogenized or discretized during representation construction, causing affinity-relevant distinctions to be lost before interaction modelling. We hypothesized that this representation-stage information loss constitutes an upstream bottleneck that cannot be reliably overcome by increasingly complex interaction predictors. To test this hypothesis, we developed ReGeoDTA, a representation-preserving framework that maintains affinity-relevant chemical heterogeneity in molecular representations and continuous geometric relationships in protein structures. Across three benchmark datasets, ReGeoDTA consistently improved affinity prediction, and the proposed representation-preserving strategies retained their benefits across diverse DTA architectures. Controlled representation degradation progressively reduced predictive performance, whereas increasing downstream predictor complexity failed to recover information lost during representation construction. These findings identify representation fidelity as an upstream design principle for accurate and generalizable drug--target affinity prediction, with potential implications for computational compound prioritization.
Yixiao Li, Yining Qian, Yefan Chen +5
College of Information Science and Engineering, Northeastern University, Shenyang 110819, China · School of Computer and Communication Engineering, Northeastern University at Qinhuangdao, Qinhuangdao 066004, China · School of Computer Science and Engineering, Northeastern University, Shenyang 110819, China +1
Drug-target affinity (DTA) prediction is widely used to prioritize candidate compounds before costly experimental screening. DTA models are often compared under a single data split, even though deployment may require extrapolation to new chemical series, new protein targets, or both. We ask whether the distribution shift used for evaluation changes which architecture appears best. We curate 718,800 unique drug-protein pairs from the ChEMBL and BindingDB datasets. We compare a Morgan-fingerprint + protein-CNN baseline with 12 controlled architectures that combine four drug representations with three ESM-2 interaction modes. Mean validation RMSE increases from 0.950 and 0.945 under scaffold and fingerprint-cluster OOD to 1.299 and 1.321 under protein-cluster and dual OOD. Model rankings are similar across the two chemical shifts (tau = 0.79), but agreement with scaffold OOD falls under protein OOD (tau = 0.39) and reverses under dual OOD (tau = -0.55). Held-out evaluation, repeated seeds, group-aware bootstrap analysis, and a size-matched control support the same conclusion: architecture selection depends on the form of extrapolation, not only on average error or training-set size. DTA benchmarks should therefore match the chemical and target shifts expected at deployment.
Minjae Chung, Clara Li, Malar Paavai Muthukumaran +8