Acta Pharmaceutica Sinica B

Home Latest Articles

Latest Articles

Editorial of special column on machine learning in drug discovery

Guobo Li, Changguo Zhan

Acta Pharmaceutica Sinica B. 2026, 16(7): 3994-3995.
Artificial intelligence in biologic drug discovery: A review of methodological evolution and therapeutic applications

Jianxin Tang, Daohong Gong, Honglin Li, Shiliang Li

Acta Pharmaceutica Sinica B. 2026, 16(7): 3996-4023.

Biologic drugs, primarily comprising proteins and nucleic acids, have emerged as powerful therapeutic modalities; however, their discovery and optimization are often hindered by their inherent complexity. The advent of artificial intelligence (AI), particularly deep learning, is catalyzing a paradigm shift in this field, transitioning it from a process reliant on serendipity and laborious experimentation to a data-driven engineering discipline. This review systematically charts the co-evolution of AI methodologies and their transformative applications across the modern biologic drug development pipeline. We first outline AI’s methodological progression, from language models deciphering biological sequence grammar to structure prediction models like AlphaFold making macromolecular folds computationally accessible, and finally to generative models enabling de novo molecular creation. We then explore the practical impact of these technologies in two core phases: the de novo design of novel biologics with bespoke functions and the subsequent multi-parameter engineering and optimization of these candidates for clinical viability. While the potential is immense, significant strategic challenges remain, including the need to build a new AI-native experimental ecosystem and bridge the profound complexity gap between molecular-level predictions and systemic in vivo outcomes. Overcoming these obstacles will usher in a new era of AI-driven, automated closed-loop drug discovery.
Machine learning reshapes the paradigm of nanomedicine research

Ziye Wei, Shijie Zhuo, Yixin Zhang, Lianlian Wu, Xiang Gao, Song He, Xiaochen Bo, Wenhu Zhou

Acta Pharmaceutica Sinica B. 2026, 16(7): 4024-4050.

Nanodrug delivery systems (NDDS) have demonstrated outstanding performance in drug delivery due to their efficient delivery capacity, targeting ability, and biocompatibility. However, the development of nanomedicines still heavily relies on the expertise of formulation scientists and extensive trial-and-error experiments. Despite the abundance of data in nanoscience, traditional biological research often struggles to effectively process, analyze, and utilize these datasets, limiting nanomedicine studies to a “one-to-one” approach. Against this backdrop, the rapid growth of artificial intelligence (AI) and machine learning (ML) offers a new paradigm for nanomedicine research. Unlike traditional statistical analyses and mathematical models, AI and ML provide deeper insights into big data, enhancing the efficiency of nanomedicine development while steering the field toward more intelligent and more precise research approaches. This review focuses on milestone studies that use ML to reshape nanomedicine research from a pharmaceutics perspective, highlighting how data-driven ML models can guide new directions in nanomedicine development.
Accurate and task-agnostic modeling of enzymatic reactions through multimodal relational learning

Acta Pharmaceutica Sinica B. 2026, 16(7): 4051-4066.

Enzymatic reactions play an emerging role in a broad spectrum of scientific and industrial applications. The inherent complexity of enzymes, such as their substrate specificity, conformational flexibility, and the vast diversity of reactions involved, poses substantial challenges for the advanced computational prediction of enzymatic reactions with desirable accuracy. Moreover, existing approaches are mostly tailored for a specific sub-task, such as substrate prediction or binding site annotation, which limits their applicability. In this study, we introduce ERAM, a task-agnostic multimodal learning framework capable of addressing a broad range of downstream applications with both accuracy and efficiency. ERAM aligns pre-trained molecular representations from Protein Language Model with the knowledge of enzyme catalysis by modeling enzymatic reactions as multi-relational data. In enzyme retrieval tasks, ERAM achieves an improvement of 28.31% in mean average precision compared with the state-of-the-art (SOTA) method, CREEP. In substrate prediction tasks, ERAM outperforms the SOTA method ESP, achieving average improvements of 35.53% and 22.97% in Matthews correlation coefficient across two datasets. Additionally, ERAM exhibits commendable interpretability by assigning higher attention weights to binding sites, resulting in lower false-positive rates (42.36%) and higher overlap scores (70.59%) in the unsupervised binding site prediction task compared to RXNAA Mapper. By learning embeddings of substrates, enzymes, and products within a unified knowledge graph latent space, ERAM demonstrates its potential as a versatile and effective tool for enzyme catalysis research.
FlowDock: A unified flow-based framework for flexible protein-ligand docking and binding affinity prediction

Acta Pharmaceutica Sinica B. 2026, 16(7): 4067-4082.

Accurate prediction of protein-ligand complexes and binding affinity is critical for hit identification and optimization for structure-based drug design. Traditional docking simulates binding processes with searching algorithms guided by energy-scoring functions, which are quite computationally expensive and time-intensive. In contrast, deep learning approaches offer a cost-effective alternative, yet often generate conformations with limited physicochemical validity and fail to account for protein flexibility. To address these pitfalls, we propose FlowDock, a multitask framework enhanced by Bayesian Flow Networks. FlowDock simultaneously generates accurate protein-ligand complex structures and predicts binding affinity while incorporating protein conformational flexibility. By leveraging multimodal intramolecular representations with a deep equivariant generative model, our method iteratively refines complex in latent space, ensuring rapid and stable generation. Benchmark evaluations demonstrate that FlowDock achieves state-of-the-art performance in binding pose prediction, especially physical plausibility, and virtual screening capability, alongside reliable binding affinity predictions. By providing deeper molecular insights into dynamic protein-ligand interactions, FlowDock represents a robust tool for accelerating the rational development of therapeutics.
Engineering selective PI3Kγ inhibitors with antitumor and immunomodulatory potentials

Acta Pharmaceutica Sinica B. 2026, 16(7): 4083-4102.

PI3Kγ represents a promising therapeutic target for its pivotal role in macrophage recruitment and polarization and its significant association with tumor invasion and metastasis. In this study, a series of novel indole-based PI3Kγ selective inhibitors were generated by machine learning combined with molecular hybridization. Intriguingly, the representative IHA-5f displayed picomolar-level potency and highly selective inhibition to PI3Kγ relative to PI3Kα/β/δ. Moreover, IHA-5f manifested prominent anti-melanoma activity in vitro and in vivo with no detectable visceral toxicity. Mechanistically, IHA-5f efficiently suppressed tumor cell proliferation and migration, and induced apoptosis by suppressing the PI3Kγ/AKT/NF-κB signaling axis. Concurrently, it restrained the M2 polarization of tumor-associated macrophages, thereby augmenting the antitumor immune response. This study underscores the potential for PI3Kγ inhibitors as immunomodulators and direct antitumor agents.
Revolutionizing drug discovery from natural products: The roles of artificial intelligence and multi-omics in accelerating innovation

Boyang Wang, Qingyuan Liu, Weibo Zhao, Tingyu Zhang, Dingfan Zhang, Chayanis Sutcharitchan, Shao Li

Acta Pharmaceutica Sinica B. 2026, 16(7): 4103-4127.

Natural products and their derivatives have long been crucial in drug therapy, especially in traditional medicine. However, challenges in screening, isolation, characterization, and optimization have slowed their development in the pharmaceutical industry. Recent advancements in artificial intelligence (AI) and multi-omics technologies are revitalizing this field. AI offers powerful tools for understanding natural compounds, enhancing molecular representations, and supporting tasks such as binding prediction, drug repurposing, and retrosynthesis. Moreover, generative models are aiding in natural product optimization and the creation of pseudo-natural compounds. At the same time, multi-omics technologies, including genomics, transcriptomics, proteomics, and metabolomics, have enabled high-throughput studies of plant traits, synthesis, regulatory mechanisms, and quality control, providing valuable data for AI model development. These advancements help accelerate the discovery of new compounds with medicinal potential. Furthermore, in the field of traditional Chinese medicine research, which is largely based on natural plant sources, AI systems exemplified by UNIQ system, combining AI and multi-omics, have been instrumental in mechanistic studies and new drug development. This study comprehensively discusses the algorithms and applications of AI and multi-omics technologies in the drug development of natural compounds and plants, as well as summarizing relevant databases which might provide high-quality data for the future development of AI algorithms targeting natural products.
Unveiling the bioactive landscape of drug inactive ingredients (DIGs) using deep transfer learning

Minjie Mou, Jinsong Zhang, Xingang Liu, Hao Yang, Tingting Fu, Hengbin Zhang, Yimiao Zhu, Tianle Niu, Xuedong Li, Yichao Ge, Ziqi Pan, Xinyu Liu, Huaicheng Sun, Tianyuan Zhang, Yang Zhang, Feng Zhu, Jianqing Gao

Acta Pharmaceutica Sinica B. 2026, 16(7): 4128-4146.

In a drug product, the major components by mass are the drug inactive ingredients (DIGs), which raises great concerns about their unwanted effects and clinical toxicities. It is demanded to unveil their proteome-wide bioactive landscape using computational methods. However, existing methods are impeded by either incapability to scan human proteome or inaccuracy in DIGs’ bioactivity prediction. Here, a cross-attention transformer model, titled TransDIG, leveraging cross-module deep transfer learning was therefore developed to map the bioactive landscape of DIGs using minimal experimental data. First, the generalizability and interpretability of this model was verified by the prediction of zero-shot proteins and identification of key atoms/residues, respectively. Then, the bioactive landscape of hundreds of DIGs was unveiled using TransDIG, and thousands of potential bioactivities were found for the DIGs currently employed in pharmaceutical industry. Finally, the bioactivities of four popular DIGs were identified based on the landscape and experimentally validated by activity assay. As a result, the colorant β-carotene was validated to inhibit a critical drug transporter, and our study presented the first in vitro evidence of the bioactivity of the antioxidant dodecyl gallate that has not previously been reported to regulate any human protein. This study might offer insights for the design of drug formulation and its clinical utilization.
Deep learning-driven discovery and mechanism of action study of a minimalist conopeptide targeting α7 nicotinic acetylcholine receptor

Acta Pharmaceutica Sinica B. 2026, 16(7): 4147-4165.

Despite extensive structural and functional characterization of the α7 nicotinic acetylcholine receptor, valuable structural insights into its interactions with conopeptides remain limited, thereby hindering the rational development of peptide-based modulators for this clinically important receptor subtype. Here, we present an integrated pipeline combining deep learning, structural biology, computational modeling and electrophysiology to accelerate the discovery and optimization of α7 nAChR-targeting conopeptides. To overcome data scarcity, we developed a deep learning model using the ESM-2 protein language framework, enabling efficient screening of 689 disulfide-poor conopeptides. This approach identified SS1, a novel antagonist of α7 nAChR, which was systematically optimized via structure-activity relationship studies to yield [△QP,S8R]SS1—a minimalist peptide with nanomolar potency (IC₅₀ = 49.2 nmol/L), enhanced selectivity, and improved stability. Cryo-EM and computational modeling resolved the 3.3 Å resolution structure of α7 nAChR bound to [S8R]SS1, revealing a unique binding mode stabilized by hydrogen bonds, hydrophobic interactions, and glycan contacts, while hybrid receptor conformations (closed/desensitized) elucidated its inhibitory mechanism. This work establishes a transformative deep learning-to-experiment framework for accelerating the discovery and optimization of nature-inspired peptide therapeutics.
Computation-driven discovery of novel chromone derivatives for the treatment of triple-negative breast cancer via mTOR-targeted inhibition and triggering antitumor immunity

Acta Pharmaceutica Sinica B. 2026, 16(7): 4166-4195.

Triple-negative breast cancer (TNBC) is a highly aggressive and heterogeneous subtype of breast cancer characterized by early metastasis, poor prognosis, and high recurrence rates. Targeting dysregulated PI3K/Akt/mTOR signaling and triggering anti-tumor immunity represent promising strategies for TNBC therapy. In this study, we report the discovery of a series of novel chromone derivatives as potent mTOR inhibitors by artificial intelligence-assisted drug design and structure-based drug design. The optimal compound, MT-44, was a highly selective mTOR inhibitor and showed no obvious binding activity to a broad panel of 200 kinases, and it exhibited nanomolar-level mTOR inhibitory and anti-TNBC cells proliferative activities. MT-44 effectively blocked the PI3K/Akt/mTOR signaling pathway and exerted robust anti-tumor efficacy in an MDA-MB-231 xenograft mouse model. Furthermore, MT-44 activated pattern recognition receptor TLR2 and upregulated the cGAS/STING signaling pathway, and reshaped the tumor microenvironment, thereby enhancing the tumor immune landscape. Collectively, our findings highlighted MT-44 as a highly selective and potent mTOR inhibitor with dual-targeted therapeutic and immunomodulatory effects, offering an appealing strategy for TNBC.

Previous 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 Next

Submission

Peer Review

Staff Login