Advances in Generative Modeling for Molecular Sciences

The field of generative modeling is rapidly advancing in the context of molecular sciences, with a focus on developing innovative methods for molecular generation, representation learning, and controllable graph generation. Recent developments have highlighted the importance of integrating biomedical knowledge graphs into generative models to improve targeted drug discovery. Additionally, there is a growing interest in designing novel architectures that can balance representation learning and generation aims, such as the use of diffusion models and information bottlenecks. Noteworthy papers in this area include the introduction of K-DREAM, a knowledge-driven embedding-augmented model that leverages knowledge graphs to augment diffusion-based generative models, and the proposal of GraphBFN, a hierarchical coarse-to-fine framework based on Bayesian Flow Networks that operates on the parameters of distributions. Other notable works include the development of TreeDiff, a Monte Carlo Tree Search guided dual-space diffusion framework for controllable graph generation, and ProteinAE, a novel protein diffusion autoencoder designed to overcome the challenges of representing protein structures.

Sources

Combined Representation and Generation with Diffusive State Predictive Information Bottleneck

Augmenting generative models with biomedical knowledge graphs improves targeted drug discovery

Hierarchical Bayesian Flow Networks for Molecular Graph Generation

Controllable Graph Generation with Diffusion Models via Inference-Time Tree Search Guidance

ProteinAE: Protein Diffusion Autoencoders for Structure Encoding

Designing ReLU Generative Networks to Enumerate Trees with a Given Tree Edit Distance

Diffusion Transformers with Representation Autoencoders

Manifold Decoders: A Framework for Generative Modeling from Nonlinear Embeddings

Contrastive Diffusion Alignment: Learning Structured Latents for Controllable Generation

Built with on top of