Chang-Yu Hsieh

LumiCharge: Spherical Harmonic Convolutional Networks for Atomic Charge Prediction in Drug Discovery.

Atomic charge is crucial in drug design for analyzing reactive sites and interactions between ligands and targets. While quantum mechanical methods offer high accuracy, they are generally computationally costly. Conversely, empirical approaches, while computationally efficient, frequently suffer from lack of precision and generalizability. Recent a number of machine learning-based models have been developed for atomic charge predictions, but they struggle with accurately representing molecular structures and capturing the chemical environments affecting atomic charges, thus limiting their generalization and accuracy. To overcome these limitations, we propose LumiCharge, a novel atomic charge prediction framework that incorporates high-order spherical harmonics convolutions and explicitly models multibody interactions. In constructing this model, we employ a strategy that integrates both high- and low-order information, enhancing its geometric spatial perception capability, which is currently underexplored in the field. Benchmark evaluations demonstrate that LumiCharge outperforms state-of-the-art (SOTA) models by 30%-60% across diverse data sets. Additionally, in cross-scale experiments, LumiCharge demonstrates exceptional extrapolation capability and robustness across molecules of varying sizes, effectively overcoming the limitations imposed by molecular sizes. On an external halogen-containing test set, LumiCharge achieves an RMSE of 0.055e, meeting practical application requirements. Finally, a case study of virtual screening for the androgen receptor (AR) target further validates its outstanding accuracy compared to the OPLS3e force field and other deep learning (DL)-based baseline models, highlighting its exceptional generalization capacity and practical utility in real-world scenarios.

PepBAN: A Deep Learning Framework with Bilinear Attention and Adversarial Learning for Peptide-Protein Interaction Prediction

Accurate prediction of the peptide-protein interaction (PepPI) is crucial for developing peptide-based therapeutics and vaccines. However, this computational task has traditionally faced significant challenges, such as the scarcity of structure data along with the corresponding label of the binding affinity for bound complexes. To address these challenges, we introduce PepBAN, a deep learning framework for modeling PepPI predictions. PepBAN incorporates two technical advancements: (1) adopting the protein language model ESM-2 to characterize proteins and ESM-2 or a graph-based foundation model for peptides without structure data and (2) leveraging the conditional domain adversarial learning to enhance generalization across a broad range of protein targets, especially when there are limited binding data. At the core of PepBAN is a bilinear attention network (BAN) that effectively learns the pattern of pairwise local interactions, enables the identification of key residues participating in the peptide-protein interactions, and offers an intuitive approach to interpret the underlying mechanisms of PepPIs via analyzing attention weights. Our numerical experiments demonstrated that PepBAN outperformed the previous state-of-the-art models across several well-established benchmark studies. Furthermore, we evaluated PepBAN's applicability in predicting cyclic peptide-protein interactions, a task that poses significant challenges due to the presence of noncanonical amino acids. These nonstandard residues require specialized handling, which most existing sequence-based PepPI prediction models did not adequately address, and we adopt an atom-resolved molecular graph approach to process cyclic peptides. Despite this complexity, PepBAN demonstrated a clear advantage by achieving a superior prediction performance and offering a distinct edge in tackling the emerging chemical space of cyclic peptides, which has great potential for novel therapeutic development. In summary, PepBAN serves as a valuable tool for advancing peptide-based drug and therapeutic development.

SynGFN: learning across chemical space with generative flow-based molecular discovery

ChargeNet: E(3) Equivariant Graph Attention Network for Atomic Charge Prediction

Atomic charge is a fundamental quantum chemical property essential for advancing drug design and discovery. Although quantum mechanics (QM) methods offer the highest level of accuracy, their computational demands scale quadratically with the number of atoms, limiting their practicality for large-scale applications. In light of this, empirical and semiempirical methods have been introduced to improve computational efficiency, albeit often at the expense of accuracy. The advent of artificial intelligence has witnessed a growing application of machine learning (ML) techniques to accelerate atomic charge predictions. However, existing ML models often suffer from low accuracy and limited generalization capabilities. To address these challenges, we introduce an advanced equivariant graph attention neural network specifically engineered to model long-range atomic electrostatic interactions with high precision. This model introduces a sophisticated global graph attention mechanism, enabling it to capture charge contributions across multiple scales. By utilizing a combination of structural symmetry-preserving transformations and multiscale attention, our approach not only preserves the inherent symmetries of molecular structures but also substantially improves the model's accuracy, generalization, and robustness in complex scenarios. Our empirical analyses demonstrate that, compared to leading baseline models, the proposed model improves charge prediction accuracy by over 40% on average across various charge-calculation schemes. Remarkably, the model achieves superior performance on the external RESP (restrained electrostatic potential) test data sets, with a 54.6% improvement over the baseline. Additionally, we evaluated our charge model under the setting of virtual screening, where it outperforms both the OPLS3 charges and baseline deep learning models across all evaluation metrics, highlighting its extensive potential for scientific discovery.

ECloudGen: leveraging electron clouds as a latent variable to scale up structure-based molecular design

MetalloDock: Decoding Metalloprotein-Ligand Interactions via Physics-Aware Deep Learning for Metalloprotein Drug Discovery.

Accurate prediction of metalloprotein-ligand interactions is critical for metalloprotein-targeted drug discovery. Conventional docking tools and existing deep learning (DL) models fail to reliably capture metal-ligand interactions, hampering the discovery of potent metalloprotein inhibitors. Here, we propose MetalloDock, the first DL-based docking framework specially designed for metalloprotein targets. By innovatively integrating an autoregressive spatial decoding engine with a physics-constrained geometric generation paradigm, MetalloDock can precisely reconstruct metal coordination geometries and accurately capture metal-ligand interactions, which enhance both the accuracy of metalloprotein-ligand docking and binding affinity prediction. Extensive evaluations on our custom-built benchmark data set demonstrate that MetalloDock outperforms existing methods, including AlphaFold3, in docking success rate and virtual screening performance for metalloprotein targets. In real-world applications, MetalloDock successfully identified multiple novel hit compounds in a virtual screening campaign targeting the prostate-specific membrane antigen. Additionally, it enabled rational drug design for acidic polymerase endonuclease, leading to the discovery of potent inhibitors. These results highlight the broad applicability of MetalloDock in accelerating metalloprotein-targeted drug discovery and provide a standardized framework for future evaluation of metalloprotein-specific docking algorithms.

A virtual platform for automated hybrid organic-enzymatic synthesis planning

The integration of organic synthesis with enzymatic catalysis offers a promising route toward efficient and sustainable construction of complex molecules. While organic synthesis enables diverse transformations, enzymatic catalysis enhances stereoselectivity under mild conditions, improving cost-effectiveness and environmental impact. However, current enzymatic synthesis planning algorithms face challenges in formulating robust hybrid organic–enzymatic strategies. Key issues include the difficulty in devising hybrid planning approaches and the reliance on template-based enzyme recommendations, which limits their adaptability across diverse scenarios. Here we show ChemEnzyRetroPlanner, an open-source hybrid synthesis planning platform that combines organic and enzymatic strategies with AI-driven decision-making. The platform features advanced computational modules, including hybrid retrosynthesis planning, reaction condition prediction, plausibility evaluation, enzymatic reaction identification, enzyme recommendation, and in silico validation of enzyme active sites. A central innovation is the RetroRollout* search algorithm, which outperforms existing tools in planning synthesis routes for organic compounds and natural products across multiple datasets. ChemEnzyRetroPlanner provides an intuitive graphical interface and programmatic APIs for scalability, while leveraging the chain-of-thought strategy and the Llama3.1 model to autonomously activate hybrid synthesis strategies for diverse scenarios. The results indicate that this fully automated, open-source system holds potential value for improving the efficiency and sustainability of molecular synthesis. The integration of organic and enzymatic synthesis enhances molecule construction efficiency. Here, the authors present ChemEnzyRetroPlanner, an AI-driven platform for automated hybrid synthesis planning, improving synthesis route efficiency and sustainability.

A scalable and quantum-accurate foundation model for biomolecular force fields via linearly tensorized quadrangle attention

Accurate atomistic biomolecular simulations are vital for understanding disease mechanisms and drug discovery, yet existing methods struggle to balance quantum-mechanical accuracy with computational scalability. Classical force fields often lack precision, while quantum methods are computationally prohibitive for complex biological systems. Here we show that LiTEN, a scalable equivariant neural network, resolves this dilemma by efficiently modeling complex three- and four-body interactions with linear complexity via Linearly Tensorized Quadrangle Attention. We introduce LiTEN-FF, a foundation model pre-trained on extensive datasets to ensure broad chemical generalization across diverse molecular spaces. We demonstrate that LiTEN achieves state-of-the-art accuracy on standard benchmarks, consistently outperforming leading approaches in both precision and speed. Furthermore, LiTEN-FF enables comprehensive modeling tasks, ranging from geometry optimization to free energy surface construction, with high computational efficiency for large biomolecules. This framework provides a physically grounded, versatile foundation for advanced biomolecular modeling and drug design applications.

mRNABERT: advancing mRNA sequence design with a universal language model and comprehensive dataset

Designing effective mRNA sequences for therapeutics remains a formidable challenge. Inspired by successes in protein design, language models (LMs) are now being applied to RNA, but progress is often impeded by the lack of comprehensive training data. Existing models are frequently limited to UTR or CDS regions, restricting their application for complete mRNA sequences. We introduce mRNABERT, a robust, all-in-one mRNA designer pre-trained on the largest available mRNA dataset. To enhance performance, we propose a dual tokenization scheme with a cross-modality contrastive learning framework to integrate semantic information from protein sequences. On a comprehensive benchmark, mRNABERT demonstrates state-of-the-art performance, outperforming previous models in the majority of tasks for 5’ UTR and CDS design, RNA-binding protein (RBP) site prediction, and full-length mRNA property prediction. It also surpasses large protein models in several related tasks. In conclusion, mRNABERT’s superior performance across these diverse tasks signifies a substantial leap forward in mRNA research and therapeutic development. Designing complete mRNA sequences for new vaccines and therapies is a complex challenge. Here, the authors develop mRNABERT, a foundational AI model that designs entire mRNA sequences and demonstrates superior performance across comprehensive benchmarks.

Token-Mol 1.0: tokenized drug design with large language models

The integration of large language models (LLMs) into drug design is gaining momentum; however, existing approaches often struggle to effectively incorporate three-dimensional molecular structures. Here, we present Token-Mol, a token-only 3D drug design model that encodes both 2D and 3D structural information, along with molecular properties, into discrete tokens. Built on a transformer decoder and trained with causal masking, Token-Mol introduces a Gaussian cross-entropy loss function tailored for regression tasks, enabling superior performance across multiple downstream applications. The model surpasses existing methods, improving molecular conformation generation by over 10% and 20% across two datasets, while outperforming token-only models by 30% in property prediction. In pocket-based molecular generation, it enhances drug-likeness and synthetic accessibility by approximately 11% and 14%, respectively. Notably, Token-Mol operates 35 times faster than expert diffusion models. In real-world validation, it improves success rates and, when combined with reinforcement learning, further optimizes affinity and drug-likeness, advancing AI-driven drug discovery. In this work the authors present Token-Mol, a token-only 3D drug design model, which deploys the Gaussian cross-entropy (GCE) loss function for regression tasks. It exhibits superior performance in molecular conformation generation, property prediction, and pocket-based generation, thus opening up new avenues for drug design.

HiCLR: Knowledge-Induced Hierarchical Contrastive Learning with Retrosynthesis Prediction Yields a Reaction Foundation Model

Reaction representation learning is of paramount importance for adopting deep-learning-based chemistry modeling to solve real-world tasks such as synthesis planning. Most prevailing models are prestrained by self-supervised objectives that rely solely on the chemical structure information. Since structurally similar reactions could possess entirely distinct properties (e.g., reaction yields) and the synthesis-related tasks are highly heterogeneous, there are inherent limitations in constructing a foundational reaction model within the existing approaches. To tackle this limitation, we propose HiCLR, a knowledge-induced hierarchical contrastive learning framework for chemical reactions, by introducing relational inductive bias to forge chemically meaningful and generally applicable reaction fingerprints. Critically, the pretraining scheme combining both retrosynthesis prediction and contrastive loss enables HiCLR to tackle generation-based and understanding-based tasks simultaneously. Comprehensive experiments demonstrate that HiCLR successfully organizes the reaction space into hierarchical global semantic clusters, aligned well with prior knowledge. Consequently, HiCLR is the first foundation model that can be broadly applied to various synthesis-related tasks, and it achieves state-of-the-art performance in reaction classification, reaction condition recommendation, reaction yield prediction, synthesis planning, and even molecular property prediction. HiCLR demonstrates clear benefits in incorporating domain knowledge to guide the learning of neural networks, expediting AI-driven advancements in chemistry.

RSGPT: a generative transformer model for retrosynthesis planning pre-trained on ten billion datapoints

Retrosynthesis planning is a crucial task in organic synthesis, and deep-learning methods have enhanced and accelerated this process. With the advancement of the emergence of large language models, the demand for data is rapidly increasing. However, available retrosynthesis data are limited to only millions. Therefore, we pioneer the utilization of the template-based algorithm to generate chemical reaction data, resulting in the production of over 10 billion reaction datapoints. A generative pretrained transformer model is subsequently developed for template-free retrosynthesis planning by pre-training on 10 billion generated data. Inspired by the strategies of large language models, we introduce reinforcement learning to capture the relationships among products, reactants, and templates more accurately. Experiments demonstrate that our model achieves state-of-the-art performance on the benchmark, with a Top-1 accuracy of 63.4%, substantially outperforming previous models. Computer-aided synthesis-planning methods have significantly assisted synthesis planning. In this work, the authors present RSGPT, a generative model pre-trained on ten billion data points, achieving state-of-the-art performance for synthesis planning