Introduces a memory unit that modifies the RL reward function to penalize previously explored chemical scaffolds, substantially increasing the diversity of generated molecules while maintaining relevance to known active ligands.
Proposes MolecularRNN, a graph recurrent model that generates molecular graphs atom-by-atom with 100% validity via valency-based rejection sampling, then shifts property distributions using policy gradient reinforcement learning.
Proposes ORGAN, a framework that extends SeqGAN with domain-specific reward functions via reinforcement learning, enabling tunable generation of molecules optimized for druglikeness, solubility, and synthesizability while maintaining sample diversity.
PharMolixFM proposes a unified framework for all-atom foundation models using three multi-modal generative approaches (diffusion, flow matching, BFN) and demonstrates competitive docking accuracy with fast inference.
Introduces a policy-based reinforcement learning method that fine-tunes an RNN pre-trained on ChEMBL SMILES to generate molecules with specified desirable properties, using an augmented episodic likelihood that anchors the agent to its prior while optimizing a user-defined scoring function.
A systematic study of SMILES augmentation strategies for molecular property prediction, showing that augmentation consistently improves CNN and RNN performance and that prediction variance across SMILES correlates with model uncertainty.
MG-BERT combines GNN-style local attention with BERT’s masked pretraining on molecular graphs, learning context-sensitive atomic representations that improve ADMET property prediction across 11 benchmark datasets.
AlphaDrug generates drug candidates for specific protein targets by combining an Lmser Transformer (with hierarchical encoder-decoder skip connections) and Monte Carlo tree search guided by docking scores, achieving higher binding affinities than known ligands on 86% of test proteins.
Proposes Augmented Hill-Climb, a hybrid RL strategy for SMILES-based generative models that improves sample efficiency ~45-fold over REINVENT by filtering low-scoring molecules from the loss computation, with diversity filters to prevent mode collapse.
Shows that divergence between optimization and control scores during goal-directed molecular generation is explained by pre-existing disagreement among QSAR models on the training distribution, not by algorithmic exploitation of model-specific biases.
BindGPT formulates 3D molecular design as autoregressive text generation over combined SMILES and XYZ tokens, using large-scale pre-training and reinforcement learning to achieve competitive pocket-conditioned molecule generation.
Winter et al. propose CDDD, a translation-based encoder-decoder that learns continuous molecular descriptors by translating between equivalent chemical representations like SMILES and InChI, pretrained on 72 million compounds.