The discovery of CAP-Gly domain-containing linker protein 1(CLIP1)-Leukocyte tyrosine kinase (LTK) as an oncogenic fusion reveals a unique dependency not only on LTK kinase activity but also on CLIP1-mediated multimerization, a noncatalytic function that drives oncogenic signaling. While this fusion is currently targeted with anaplastic lymphoma kinase inhibitors, their exclusive focus on kinase inhibition leaves the scaffolding function intact, necessitating a complete protein clearance strategy. Here, we report the AI-guided development of a first-in-class proteolysis-targeting chimera (PROTAC) designed to selectively degrade the CLIP1-LTK fusion protein. By integrating deep learning models for ternary complex prediction with structure-based molecular optimization, we designed DCL05, an orally bioavailable degrader of CLIP1-LTK fusion protein, achieving picomolar degradation potency (DC50 = 40 pM) and robust antitumor activity. DCL05 consistently outperformed existing kinase inhibitors across a broad spectrum of LTK resistance-associated mutations, both in vitro and in vivo. Collectively, our study explores resistance-associated contexts of LTK and establishes a structure-guided PROTAC development pipeline, providing a promising therapeutic strategy for overcoming acquired resistance in kinase-driven cancers.
Shicheng Chen, Haiting Duan, S. Zhong et al.· Proceedings of the National...· 0 citations
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.
Accurate prediction of blood-brain barrier permeability (BBBP) is essential for central nervous system drug discovery, yet existing models are often limited by their reliance on predefined physicochemical descriptors, small-molecule-centered training sets, or conformation-dependent representations, which restricts their transferability across chemically diverse modalities especially peptides. In addition, publicly available BBBP datasets remain fragmented, inconsistently standardized, and weakly controlled for molecular redundancy, increasing the risk of data leakage and overestimated model performance. In this study, we propose BBBP-Atlas, a structure-aware BBB permeability prediction model designed for unified modeling of small molecules and peptides with the first cross-modal dataset OmniBBBP. Designed to bypass descriptor and conformation dependencies, our model represents standardized molecular structures as atom-level graphs to capture local atom-bond environments and long-range topological dependencies associated with BBB transport. This design enables direct learning of structure-permeability relationships from molecular topology. For model training and evaluation, we curated a cross-modal, redundancy-filtered database OmniBBBP that seamlessly unifies small molecules and complex peptides, containing 10,218 unique compounds with 9,316 small molecules and 902 peptides. BBBP-Atlas achieved an accuracy of 0.8914 and an MCC of 0.7678 on the independent test set. On a balanced external benchmark of 200 compounds, our model reached an AUC of 0.9108, an accuracy of 0.8500, and an MCC of 0.7000, outperforming LightBBB by an absolute MCC gain of 6%. Case studies further showed that BBBP-Atlas captured clinically meaningful BBB permeability patterns, correctly identifying lorlatinib as BBB-permeable and vancomycin as BBB-impermeable with high confidence. The OmniBBBP-backed BBBP-Atlas offers a versatile and cross-modal approach for single-compound prediction, batch screening, and dataset exploration for CNS drug discovery. BBBP-Atlas is available at https://cadd.drugflow.com/bbbp/.
Xin Shen, Qun Su, Hao Luo et al.· bioRxiv· 0 citations
Targeting the intrinsically disordered N-terminal domain of the androgen receptor (AR-NTD) represents a promising strategy to overcome resistance in prostate cancer. However, its inherent lack of a stable tertiary structure and highly dynamic conformational ensemble pose formidable challenges for rational drug design. This study introduces an integrated computational workflow that combines enhanced sampling techniques and machine learning collective variables to identify druggable conformations of the AR-NTD and elucidate the binding mechanism of its modulator, EPI-002. We characterize nine metastable states of the Tau-5 region and reveal that ligand recognition is driven by π–π stacking and structured water-mediated hydrogen bonds. Leveraging these insights, we perform structure-based virtual screening based on the identified druggable conformations and identify K53, a rationally designed AR-NTD antagonist, which exhibits potent anti-proliferative activity in enzalutamide-resistant prostate cancer cells. K53 directly binds the AR-NTD, suppresses AR transcriptional activity, and demonstrates high selectivity for cancer cells. This work provides a rational design paradigm for targeting intrinsically disordered proteins and offers a therapeutic candidate for resistant prostate cancer. In this work, the authors develop a machine learning–based enhanced sampling workflow to target the intrinsically disordered AR-NTD, identifying druggable conformations and enabling transferable modeling of ligand binding for rational drug discovery.
Kai Zhu, Huating Wang, Jintu Zhang et al.· Nature Communications· 0 citations
Reach audiences
Advertise in front of researchers, engineers, and readers.
High-speed wireless communication systems underpin the data-intensive demands of contemporary 5G deployments and the emerging 6G paradigm, yet sustaining high throughput, low latency, and dependable connectivity in channels that shift rapidly remains an open engineering challenge. This paper presents an ML-driven framework that combines a deep reinforcement learning (DRL)-based scheduler with a hybrid CNN-LSTM channel predictor to jointly optimize radio resource allocation, modulation order, transmit power, and bandwidth assignment in real-time. Evaluated across full-buffer, bursty, mixed-traffic, and high-mobility scenarios using a 3GPP TR 38.901-compliant simulator, the proposed scheduler achieved 23.0% higher throughput, 35.0% lower latency, and 18.1% higher energy efficiency than Proportional Fair scheduling, while reducing Quality of Service (QoS) violations by 75.9% and packet loss by 76.3%, with a Jain's Fairness Index of 0.887. The companion CNN-LSTM channel predictor achieved 94.7% prediction accuracy and a normalized mean squared error of −16.2 dB, reducing CSI-feedback overhead by 41.3% relative to reactive CSI feedback schemes and outperforming standalone CNN and LSTM baselines by 28% under high-mobility conditions. These results indicate that coupling predictive channel-state modeling with multi-objective reinforcement learning offers a practical and quantifiable pathway toward achieving next-generation wireless performance targets.
This paper investigates the application of a self-adaptive quantum simulation optimization algorithm to enhance the accuracy and efficiency of quantum simulations. Traditional simulation methods often rely on fixed parameters, limiting the potential for optimization. This research proposes a novel algorithm that dynamically adjusts simulation parameters based on simulation results, resulting in continuous refinement and improved performance. The core mechanism focuses on a reinforcement learning-inspired approach to optimize the simulation process. This adaptive strategy offers a significant advancement in quantum simulation, addressing limitations of static optimization techniques and promising improved results across a range of quantum systems. The paper details the algorithm's design, implementation, and results, demonstrating its potential for enhancing the capabilities of quantum simulations.
Jincheng Zhang· Zenodo (CERN European Organi...· 0 citations
This paper explores the application of geometric constraints within reinforcement learning to enhance policy stability and effectiveness. Traditional reinforcement learning methods often struggle with complex environments where geometric considerations are crucial. We propose a novel geometric constraint formulation that explicitly incorporates geometric relationships between states and actions, leading to improved agent behavior. The core mechanism involves representing the environment as a geometric space and leveraging geometric constraints to guide the agent's decision-making process. We demonstrate the effectiveness of this approach through a series of simulations, showcasing significant improvements in convergence speed and robustness compared to standard reinforcement learning algorithms.
Jincheng Zhang· Zenodo (CERN European Organi...· 0 citations
This paper introduces the Dynamic Semantic Embedding Network (DSE-Net), a novel approach to understanding evolving data streams. The core claim is that by integrating semantic embeddings with dynamic graph neural networks, we can achieve continuous, context-aware understanding, overcoming the limitations of static embedding models. DSE-Net employs a multi-layered architecture: a Transformer encoder for initial semantic embedding generation, a dynamic graph neural network (GNN) to model relationships within the data stream, and a reinforcement learning (RL) module to optimize the GNN's structure and parameters adaptively. Crucially, the embedding itself is updated based on the GNN's output, creating an evolving semantic representation. This approach addresses the shortcomings of existing methods, which either rely on static embeddings or static GNNs, by providing a dynamic and learning system capable of adapting to changing semantic relationships. The key innovation lies in the synergistic combination of these techniques, leading to a more robust and nuanced understanding of dynamic data. ---
Jincheng Zhang· Zenodo (CERN European Organi...· 0 citations
This paper presents a novel approach to intelligent robot control by leveraging the fusion of multiple modalities – visual, tactile, and auditory – of information. The core idea is to enhance robot adaptability and intelligence in complex environments through sophisticated multi-modal data processing and intelligent control strategy learning. We employ deep learning techniques for the initial feature extraction and fusion from each modality, followed by reinforcement learning to train the robot's control policy. The proposed system aims to achieve a higher level of robot perception and action, bridging the gap between raw sensory input and effective robotic behavior. The system is designed for adaptability to varying environmental conditions and task requirements. The key contribution lies in the integrated architecture and the utilization of deep learning for robust multi-modal feature representation and reinforcement learning for adaptive control. The efficacy of the approach is demonstrated through a theoretical framework and a conceptual design.
Jincheng Zhang· Zenodo (CERN European Organi...· 0 citations
This paper presents the Adaptive Error Correction for Complex Parameter Space (AECPS) algorithm, a novel approach to parameter estimation designed for complex parameter spaces. Traditional parameter estimation methods frequently struggle to adapt to dynamic changes in the parameter space, leading to suboptimal performance. AECPS leverages reinforcement learning to dynamically adjust error correction parameters, optimizing both accuracy and computational efficiency. We detail the core mechanism, including the reinforcement learning framework, error estimation process, and performance evaluation. The goal is to provide a robust and adaptable solution for parameter estimation in challenging environments, addressing limitations inherent in current techniques.
Jincheng Zhang· Zenodo (CERN European Organi...· 0 citations
This paper presents a novel dynamic topology optimization network routing algorithm designed to enhance network performance and resilience. The core concept involves real-time adaptation of the network topology based on dynamic network conditions such as congestion, latency, and node failures. This is achieved through the integration of reinforcement learning or evolutionary algorithms to construct routing algorithms that can intelligently adjust to network changes and support adaptive topology modifications. Unlike traditional static routing protocols, this approach allows for a continuous optimization process, leading to improved network efficiency and enhanced fault tolerance. The algorithm's effectiveness is predicated on the ability to accurately assess network state and to strategically modify the network topology to mitigate negative impacts and exploit available resources. The presented framework offers a significant advancement over existing routing methods, particularly in complex and rapidly evolving network environments.
Jincheng Zhang· Zenodo (CERN European Organi...· 0 citations
This paper presents a novel approach to neural network architecture search (NAS) leveraging meta-learning and Bayesian optimization. Traditional NAS methods often suffer from high computational costs associated with exhaustive or reinforcement learning-based exploration of the architecture search space. Our method, Bayesian Optimization with a Gaussian Process surrogate model, offers a significantly more efficient alternative. We learn from previous architecture search trials, using this knowledge to guide the selection of promising architectures in subsequent searches. This meta-learning framework allows us to rapidly converge on high-performing architectures, reducing the overall search time while maintaining competitive accuracy. The core of our approach lies in the use of a Gaussian Process to model the performance of different neural network architectures, and then employing an acquisition function to intelligently guide the exploration of the architecture space. This paper details the formulation of the problem, the implementation of the Bayesian optimization algorithm, and demonstrates its effectiveness through theoretical analysis and a discussion of the key components.
Jincheng Zhang· Zenodo (CERN European Organi...· 0 citations
A new machine-learning framework aims to improve the success rate of computational protein design while moving away from results that reproduce sequences found in nature.
MIT News · Artificial Intelligence· news.mit.eduAug 24, 2026
A new method for surgically removing training examples from a model reveals that as datasets grow, the link between what a model learns and what it produces dissolves.
