Protein language models (PLMs) have emerged as powerful tools for controllable biomolecular design, yet their post-training adaptation typically relies on costly wet-lab validation or curated preference datasets. To overcome this supervision bottleneck, we introduce unsupervised reward optimization of PLMs, a comprehensive framework for steerable protein generation without ground-truth labels. Our key insight is that task-agnostic rewards, which combine intrinsic model uncertainty with extrinsic semantic consistency informed by protein representation models, exhibit strong correlation with controllability measures across base models and temperature regimes. Building upon this discovery, we propose two offline algorithms: Soft Reward Optimization (SRO) and Binarized Reward Optimization (BRO), which effectively maximize the classical RLHF objective induced by these proxy rewards. Extensive experiments on compositional out-of-distribution prompts demonstrate that both methods significantly outperform competitive baselines (DPO, KTO), while approaching oracle performance across multiple sampling temperatures, model scales and protein families. Moreover, PLMs fine-tuned with unsupervised rewards can achieve consistently higher coverage compared to their base model in pass@k evaluations. By enabling self-improvement of PLMs through their own generated experience, our framework provides a scalable pathway toward controllable biomolecular design in settings where labeled preferences or experimental feedback are scarce or unavailable.

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.

The development of AI-assisted chemical synthesis tools requires comprehensive datasets covering diverse reaction types, yet current high-throughput experimental (HTE) approaches are expensive and limited in scope. Chemical literature represents a vast, underexplored data source containing thousands of reactions published annually. However, extracting reaction information from literature faces significant challenges including varied writing styles, complex coreference relationships, and multimodal information presentation. This paper proposes ChemMiner, a novel end-to-end framework leveraging multiple agents powered by large language models (LLMs) to extract high-fidelity chemical data from literature. ChemMiner incorporates three specialized agents: a text analysis agent for coreference mapping, a multimodal agent for non-textual information extraction, and a synthesis analysis agent for data generation. Furthermore, we developed a comprehensive benchmark with expert-annotated chemical literature to evaluate both extraction efficiency and precision. Experimental results demonstrate reaction identification rates comparable to human chemists while significantly reducing processing time, with high accuracy, recall, and F1 scores. Our open-sourced benchmark facilitates future research in chemical literature data mining.

Molecular representation learning (MRL) has shown promise in accelerating drug development by predicting chemical properties. However, imperfectly annotation among datasets pose challenges in model design and explainability. In this work, we formulate molecules and corresponding properties as a hypergraph, extracting three key relationships: among properties, molecule-to-property, and among molecules, and developed a unified and explainable multi-task MRL framework, OmniMol. It integrates a task-related meta-information encoder and a task-routed mixture of experts (t-MoE) backbone to capture correlations among properties and produce task-adaptive outputs. To capture underlying physical principles among molecules, we implement an innovative SE(3)-encoder for physical symmetry, applying equilibrium conformation supervision, recursive geometry updates, and scale-invariant message passing to facilitate learning-based conformational relaxation. OmniMol achieves state-of-the-art performance in properties prediction, reaches top performance in chirality-aware tasks, demonstrates explainability for all three relations, and shows effective performance in practical applications. Our code is available in our https://github.com/bowenwang77/OmniMol public repository.

Recently, reinforcement learning (RL) has become a common choice in enhancing the reasoning capabilities of vision-language models (VLMs). Considering existing RL-based finetuning methods, entropy intervention turns out to be an effective way to benefit exploratory ability, thereby improving policy performance. Notably, most existing studies intervene in entropy by simply controlling the update of specific tokens during policy optimization of RL. They ignore the entropy intervention during the RL sampling that can boost the performance of GRPO by improving the diversity of responses. In this paper, we propose Selective-adversarial Entropy Intervention, namely SaEI, which enhances policy entropy by distorting the visual input with the token-selective adversarial objective coming from the entropy of sampled responses. Specifically, we first propose entropy-guided adversarial sampling (EgAS) that formulates the entropy of sampled responses as an adversarial objective. Then, the corresponding adversarial gradient can be used to attack the visual input for producing adversarial samples, allowing the policy model to explore a larger answer space during RL sampling. Then, we propose token-selective entropy computation (TsEC) to maximize the effectiveness of adversarial attack in EgAS without distorting factual knowledge within VLMs. Extensive experiments on both in-domain and out-of-domain datasets show that our proposed method can greatly improve policy exploration via entropy intervention, to boost reasoning capabilities. Code will be released once the paper is accepted.

Offline meta reinforcement learning (OMRL) has emerged as a promising approach for interaction avoidance and strong generalization performance by leveraging pre-collected data and meta-learning techniques. Previous context-based approaches predominantly rely on the intuition that alternating optimization between the context encoder and the policy can lead to performance improvements, as long as the context encoder follows the principle of maximizing the mutual information between the task variable $M$ and its latent representation $Z$ ($I(Z;M)$) while the policy adopts the standard offline reinforcement learning (RL) algorithms conditioning on the learned task representation.Despite promising results, the theoretical justification of performance improvements for such intuition remains underexplored.Inspired by the return discrepancy scheme in the model-based RL field, we find that the previous optimization framework can be linked with the general RL objective of maximizing the expected return, thereby explaining performance improvements. Furthermore, after scrutinizing this optimization framework, we observe that the condition for monotonic performance improvements does not consider the variation of the task representation. When these variations are considered, the previously established condition may no longer be sufficient to ensure monotonicity, thereby impairing the optimization process.We name this issue \underline{task representation shift} and theoretically prove that the monotonic performance improvements can be guaranteed with appropriate context encoder updates.We use different settings to rein in the task representation shift on three widely adopted training objectives concerning maximizing $I(Z;M)$ across different data qualities.Empirical results show that reining in the task representation shift can indeed improve performance.Our work opens up a new avenue for OMRL, leading to a better understanding between the task representation and performance improvements.

Target-aware drug discovery has greatly accelerated the drug discovery process to design small-molecule ligands with high binding affinity to disease-related protein targets. Conditioned on targeted proteins, previous works utilize various kinds of deep generative models and have shown great potential in generating molecules with strong protein-ligand binding interactions. However, beyond binding affinity, effective drug molecules must manifest other essential properties such as high drug-likeness, which are not explicitly addressed by current target-aware generative methods. In this article, aiming to bridge the gap of multi-objective target-aware molecule generation in the field of deep learning-based drug discovery, we propose ParetoDrug, a Pareto Monte Carlo Tree Search (MCTS) generation algorithm. ParetoDrug searches molecules on the Pareto Front in chemical space using MCTS to enable synchronous optimization of multiple properties. Specifically, ParetoDrug utilizes pretrained atom-by-atom autoregressive generative models for the exploration guidance to desired molecules during MCTS searching. Besides, when selecting the next atom symbol, a scheme named ParetoPUCT is proposed to balance exploration and exploitation. Benchmark experiments and case studies demonstrate that ParetoDrug is highly effective in traversing the large and complex chemical space to discover novel compounds with satisfactory binding affinities and drug-like properties for various multi-objective target-aware drug discovery tasks. A multi-objective molecule generation method with Pareto MCTS as searching procedure in chemical space and the pretrained molecule generative model conditioned on protein targets as searching guidance to synchronously optimize multiple properties.

Given a graph G, the network collapse problem (NCP) selects a vertex subset S of minimum cardinality from G such that the difference in the values of a given measure function f(G)−f(G∖S) is greater than a predefined collapse threshold. Many graph analytic applications can be formulated as NCPs with different measure functions, which often pose a significant challenge due to their NP-hard nature. As a result, traditional greedy algorithms, which select the vertex with the highest reward at each step, may not effectively find the optimal solution. In addition, existing learning-based algorithms do not have the ability to model the sequence of actions taken during the decision-making process, making it difficult to capture the combinatorial effect of selected vertices on the final solution. This limits the performance of learning-based approaches in non-submodular NCPs.To address these limitations, we propose a unified framework called DT-NC, which adapts the Decision Transformer to the Network Collapse problems. DT-NC takes into account the historical actions taken during the decision-making process and effectively captures the combinatorial effect of selected vertices. The ability of DT-NC to model the dependency among selected vertices allows it to address the difficulties caused by the non-submodular property of measure functions in some NCPs effectively. Through extensive experiments on various NCPs and graphs of different sizes, we demonstrate that DT-NC outperforms the state-of-the-art methods and exhibits excellent transferability and generalizability.

As a significant task of pharmaceutical and chemical engineering, molecular retrosynthesis aims at predicting candidate reactants from predefined products. Treating this challenging task as a conditional generative modeling problem, we propose a hierarchical graph autoregression (HGAR) model and its pretraining-assisted multi-task learning paradigm, leading to an effective semi-template molecular retrosynthesis method. Given a product, we first construct a hierarchical graph by connecting the junction tree of its motifs to the atom-level molecular graph. Our HGAR model embeds the hierarchical graph in the motif and atom levels, respectively. The atom-level embeddings are applied to predict reaction centers and derive synthons from the product. The motif-level embeddings are applied to predict motifs and complete the corresponding synthons autoregressive, leading to the target reactants. We first pretrain the model on PCQM4M-LSC and then fine-tune it on the USPTO retrosynthesis datasets, leading to a model with good generalization power. Experiments show that our HGAR outperforms many representative molecular retrosynthesis methods, especially those semi-template ones, indicating its feasibility and effectiveness in practice.

We consider Imitation Learning with dynamics variation between the expert demonstration (source domain) and the environment (target domain). Based on the popular framework of Adversarial Imitation Learning, we propose a novel algorithm – Dyn amics Adapted Imitation Learning (DYNAIL), which incorporates the dynamics variation into the state-action occupancy measure matching as a regularization term. The dynamics variation is modeled by a pair of classifiers to distinguish between source dynamics and target dynamics. Theoreti-cally, we provide an upper bound on the divergence between the learned policy and expert demonstrations in the source domain . Our error bound only depends on the expectation of the discrepancy between the source and target dynamics for the optimal policy in the target domain . The experiment evaluation validates that our method achieves superior results on high dimensional continuous control tasks, compared to existing imitation learning methods.

As a marriage between offline RL and meta-RL, the advent of offline meta-reinforcement learning (OMRL) has shown great promise in enabling RL agents to multi-task and quickly adapt while acquiring knowledge safely. Among which, context-based OMRL (COMRL) as a popular paradigm, aims to learn a universal policy conditioned on effective task representations. In this work, by examining several key milestones in the field of COMRL, we propose to integrate these seemingly independent methodologies into a unified framework. Most importantly, we show that the pre-existing COMRL algorithms are essentially optimizing the same mutual information objective between the task variable M and its latent representation Z by implementing various approximate bounds. Such theoretical insight offers ample design freedom for novel algorithms. As demonstrations, we propose a supervised and a self-supervised implementation of I(Z; M), and empirically show that the corresponding optimization algorithms exhibit remarkable generalization across a broad spectrum of RL benchmarks, context shift scenarios, data qualities and deep learning architectures. This work lays the information theoretic foundation for COMRL methods, leading to a better understanding of task representation learning in the context of reinforcement learning.

Subpopulation shift exists widely in many real-world applications, which refers to the training and test distributions that contain the same subpopulation groups but with different subpopulation proportions. Ignoring subpopulation shifts may lead to significant performance degradation and fairness concerns. Importance reweighting is a classical and effective way to handle the subpopulation shift. However, recent studies have recognized that most of these approaches fail to improve the performance especially when applied to over-parameterized neural networks which are capable of fitting any training samples. In this work, we propose a simple yet practical framework, called reweighted mixup (RMIX), to mitigate the overfitting issue in over-parameterized models by conducting importance weighting on the ”mixed” samples. Benefiting from leveraging reweighting in mixup, RMIX allows the model to explore the vicinal space of minority samples more, thereby obtaining more robust model against subpopulation shift. When the subpopulation memberships are unknown, the training-trajectories-based uncertainty estimation is equipped in the proposed RMIX to flexibly characterize the subpopulation distribution. We also provide insightful theoretical analysis to verify that RMIX achieves better generalization bounds over prior works. Further, we conduct extensive empirical studies across a wide range of tasks to validate the effectiveness of the proposed method.

AI-aided drug discovery (AIDD) is gaining popularity due to its potential to make the search for new pharmaceuticals faster, less expensive, and more effective. Despite its extensive use in numerous fields (e.g., ADMET prediction, virtual screening), little research has been conducted on the out-of-distribution (OOD) learning problem with noise. We present DrugOOD, a systematic OOD dataset curator and benchmark for AIDD. Particularly, we focus on the drug-target binding affinity prediction problem, which involves both macromolecule (protein target) and small-molecule (drug compound). DrugOOD offers an automated dataset curator with user-friendly customization scripts, rich domain annotations aligned with biochemistry knowledge, realistic noise level annotations, and rigorous benchmarking of SOTA OOD algorithms, as opposed to only providing fixed datasets. Since the molecular data is often modeled as irregular graphs using graph neural network (GNN) backbones, DrugOOD also serves as a valuable testbed for graph OOD learning problems. Extensive empirical studies have revealed a significant performance gap between in-distribution and out-of-distribution experiments, emphasizing the need for the development of more effective schemes that permit OOD generalization under noise for AIDD.

Graph contrastive learning (GCL) has attracted a surge of attention due to its superior performance for learning node/graph representations without labels. However, in practice, the underlying class distribution of unlabeled nodes for the given graph is usually imbalanced. This highly imbalanced class distribution inevitably deteriorates the quality of learned node representations in GCL. Indeed, we empirically find that most state-of-the-art GCL methods cannot obtain discriminative representations and exhibit poor performance on imbalanced node classification. Motivated by this observation, we propose a principled GCL framework on Imbalanced node classification (ImGCL), which automatically and adaptively balances the representations learned from GCL without labels. Specifically, we first introduce the online clustering based progressively balanced sampling (PBS) method with theoretical rationale, which balances the training sets based on pseudo-labels obtained from learned representations in GCL. We then develop the node centrality based PBS method to better preserve the intrinsic structure of graphs, by upweighting the important nodes of the given graph. Extensive experiments on multiple imbalanced graph datasets and imbalanced settings demonstrate the effectiveness of our proposed framework, which significantly improves the performance of the recent state-of-the-art GCL methods. Further experimental ablations and analyses show that the ImGCL framework consistently improves the representation quality of nodes in under-represented (tail) classes.

Recent AI research plots a promising future of automatic chemical reactions within the chemistry society. This study proposes Chemist-X, a comprehensive AI agent that automates the reaction condition optimization (RCO) task in chemical synthesis with retrieval-augmented generation (RAG) technology and AI-controlled wet-lab experiment executions. To begin with, as an emulation on how chemical experts solve the RCO task, Chemist-X utilizes a novel RAG scheme to interrogate available molecular and literature databases to narrow the searching space for later processing. The agent then leverages a computer-aided design (CAD) tool we have developed through a large language model (LLM) supervised programming interface. With updated chemical knowledge obtained via RAG, as well as the ability in using CAD tools, our agent significantly outperforms conventional RCO AIs confined to the fixed knowledge within its training data. Finally, Chemist-X interacts with the physical world through an automated robotic system, which can validate the suggested chemical reaction condition without human interventions. The control of the robotic system was achieved with a novel algorithm we have developed for the equipment, which relies on LLMs for reliable script generation. Results of our automatic wet-lab experiments, achieved by fully LLM-supervised end-to-end operation with no human in the lope, prove Chemist-X's ability in self-driving laboratories.

Protein-Protein Interactions (PPIs) are fundamental means of functions and signalings in biological systems. The massive growth in demand and cost associated with experimental PPI studies calls for computational tools for automated prediction and understanding of PPIs. Despite recent progress, in silico methods remain inadequate in modeling the natural PPI hierarchy. Here we present a double-viewed hierarchical graph learning model, HIGH-PPI, to predict PPIs and extrapolate the molecular details involved. In this model, we create a hierarchical graph, in which a node in the PPI network (top outside-of-protein view) is a protein graph (bottom inside-of-protein view). In the bottom view, a group of chemically relevant descriptors, instead of the protein sequences, are used to better capture the structure-function relationship of the protein. HIGH-PPI examines both outside-of-protein and inside-of-protein of the human interactome to establish a robust machine understanding of PPIs. This model demonstrates high accuracy and robustness in predicting PPIs. Moreover, HIGH-PPI can interpret the modes of action of PPIs by identifying important binding and catalytic sites precisely. Overall, "HIGH-PPI [ https://github.com/zqgao22/HIGH-PPI ]" is a domain-knowledge-driven and interpretable framework for PPI prediction studies.

How to effectively represent molecules is a long-standing challenge for molecular property prediction and drug discovery. Recently, accumulative unlabelled molecule data have spurred the rapid development of pre-training methods for molecular representation learning. However, these works mainly focus on devising self-supervised learning tasks and/or introducing 3D geometric information based on molecular structures with little chemical domain knowledge involved. To address this issue, we propose a novel method (MolKD) by Distilling pre-trained Knowledge in chemical reactions to assist Molecular property prediction. Specifically, MolKD first learns effective representations by incorporating reaction yields to measure transformation efficiency of the reactant-product pair when pre-training on reactions. Next, MolKD introduces the reaction-to-molecule distillation to transfer cross-modal knowledge between pre-training chemical reaction data and the downstream molecular property prediction tasks. Extensive experiments show that our method can learn effective molecular representations, achieving superior performance compared with state-of-the-art baselines, e.g., 2.8% absolute Hit@1 gain on USPTO in chemical reaction prediction and 1.6% absolute AUC-ROC gain on Tox21 with 1/3 pre-training data size in molecular property prediction. Further investigations on pre-trained molecular representations indicate that MolKD learns to distinguish chemically meaningful molecular similarities, which enables molecular property prediction with high robustness and interpretability.

Graph neural networks (GNNs) are popular weapons for modeling relational data. Existing GNNs are not specified for attribute-incomplete graphs, making missing attribute imputation a burning issue. Until recently, many works notice that GNNs are coupled with spectral concentration, which means the spectrum obtained by GNNs concentrates on a local part in spectral domain, e.g., low-frequency due to oversmoothing issue. As a consequence, GNNs may be seriously flawed for reconstructing graph attributes as graph spectral concentration tends to cause a low imputation precision. In this work, we present a regularized graph autoencoder for graph attribute imputation, named MEGAE, which aims at mitigating spectral concentration problem by maximizing the graph spectral entropy. Notably, we first present the method for estimating graph spectral entropy without the eigen-decomposition of Laplacian matrix and provide the theoretical upper error bound. A maximum entropy regularization then acts in the latent space, which directly increases the graph spectral entropy. Extensive experiments show that MEGAE outperforms all the other state-of-the-art imputation methods on a variety of benchmark datasets.

Multi-goal reinforcement learning has emerged as a powerful approach for planning and robot manipulation tasks, but it faces challenges such as sparse rewards and sample inefficiency. Hindsight Experience Replay (HER) has been proposed as a solution to these challenges by relabeling goals, but it still requires a large number of samples and significant computation. To address these issues, we propose Multi-step Hindsight Experience Replay (MHER), which incorporates multi-step relabeling to improve sample efficiency. Despite the advantages of $n -$step relabeling, we theoretically and experimentally prove the off-policy $n -$step bias introduced by $n -$step relabeling may lead to poor performance in many environments. To address this issue, two bias-reduced MHER algorithms, MHER $( \lambda )$ and Model-based MHER (MMHER) are presented. MHER $( \lambda )$ exploits the $\lambda$ return while MMHER benefits from model-based value expansions. Experimental results on numerous multi-goal robotic tasks show that our solutions can successfully alleviate the off-policy $n -$step bias and achieve significantly higher sample efficiency than previous multi-goal RL baselines with little additional computation beyond HER.

It's widely acknowledged that deep learning models with flatter minima in its loss landscape tend to generalize better. However, such property is under-explored in deep long-tailed recognition (DLTR), a practical problem where the model is required to generalize equally well across all classes when trained on highly imbalanced label distribution. In this paper, through empirical observations, we argue that sharp minima are in fact prevalent in deep long-tailed models, whereas naï ve integration of existing flattening operations into long-tailed learning algorithms brings little improvement. Instead, we propose an effective two-stage sharpness-aware optimization approach based on the decoupling paradigm in DLTR. In the first stage, both the feature extractor and classifier are trained under parameter perturbations at a class-conditioned scale, which is theoretically motivated by the characteristic radius of flat minima under the PAC-Bayesian framework. In the second stage, we generate adversarial features with class-balanced sampling to further robustify the classifier with the backbone frozen. Extensive experiments on multiple long-tailed visual recognition benchmarks show that, our proposed Class-Conditional Sharpness-Aware Minimization (CC-SAM), achieves competitive performance compared to the state-of-the-arts. Code is available at https://github.com/zzpustc/CC-SAM.