The field of biological sequence generation and protein design is rapidly advancing, with a focus on developing innovative methods for controllable de novo sequence generation and protein design. Recent work has centered around improving the scalability and efficiency of generative models, enabling the design of high-quality, diverse sequences and proteins. Notably, the integration of multiple modalities, such as discrete molecular graphs and continuous 3D coordinates, has emerged as a key challenge and opportunity for advancing the field. Furthermore, the development of novel frameworks that can unify the representations of different molecules and leverage cross-domain transferability is showing promise. Some noteworthy papers in this area include:

• Gumbel-Softmax Flow Matching with Straight-Through Guidance for Controllable Biological Sequence Generation, which introduces a novel framework for controllable de novo sequence generation.
• UniMoMo: Unified Generative Modeling of 3D Molecules for De Novo Binder Design, which presents a unified framework for designing binders of multiple molecular domains using a single model.
• TransDiffSBDD: Causality-Aware Multi-Modal Structure-Based Drug Design, which proposes an integrated framework combining autoregressive transformers and diffusion models for structure-based drug design.
• CMADiff: Cross-Modal Aligned Diffusion for Controllable Protein Generation, which enables controllable protein generation by aligning physicochemical properties of protein sequences with text-based descriptions through a latent diffusion process.