The field of molecular representation learning and chemical discovery is rapidly advancing, with a focus on developing innovative methods for predicting molecular properties, identifying potential drug targets, and designing new molecules. Recent research has highlighted the importance of integrating structural and semantic information to improve predictive accuracy and interpretability. Notably, the use of generative models, graph neural networks, and multimodal learning has shown significant promise in addressing key challenges in the field, such as data scarcity and limited generalizability. Furthermore, the development of benchmarking frameworks and standardized evaluation protocols is essential for facilitating comparisons between different methods and driving progress in the field. Some noteworthy papers in this area include: PolyConFM, which introduces a conformation-centric generative foundation model for polymer modeling and design, demonstrating state-of-the-art performance on diverse downstream tasks. AtomBench presents a systematic benchmark of generative atomic structure models, providing a comprehensive evaluation of their performance on materials datasets. ScaffAug proposes a scaffold-aware generative augmentation and reranking framework for enhanced virtual screening, addressing challenges such as class imbalance and structural imbalance. Atom-anchored LLMs demonstrate the potential of large language models for molecular reasoning and retrosynthesis, achieving high success rates in identifying chemically plausible reaction sites and reactants. Copy-Augmented Representation introduces a novel molecular representation that effectively captures structural invariance in chemical reactions, leading to significant improvements in retrosynthesis prediction accuracy. ProtoMol presents a prototype-guided multimodal framework for molecular property prediction, enabling fine-grained integration and consistent semantic alignment between molecular graphs and textual descriptions. 3D-GSRD proposes a 3D molecular graph auto-encoder with selective re-mask decoding, achieving strong downstream performance and setting a new state-of-the-art on the MD17 molecular property prediction benchmark. Chem-R introduces a generalizable chemical reasoning model that emulates the deliberative processes of chemists, achieving state-of-the-art performance on comprehensive benchmarks and surpassing leading large language models. A Standardized Benchmark for Machine-Learned Molecular Dynamics presents a modular benchmarking framework for evaluating protein MD methods using enhanced sampling analysis, enabling fast and efficient exploration of protein conformational space. CDI-DTI proposes a cross-domain interpretable framework for DTI prediction, integrating multi-modal features and ensuring robust performance across different domains and in cold-start scenarios. MolBridge introduces a novel atom-level joint graph refinement framework for robust DDI event prediction, effectively modeling inter-drug associations and achieving superior performance across long-tail and inductive scenarios. MS-BART presents a unified modeling framework that maps mass spectra and molecular structures into a shared token vocabulary, enabling cross-modal learning and achieving state-of-the-art performance on key metrics.