Ambient Proteins - Training Diffusion Models on Noisy Structures

We present Ambient Protein Diffusion, a framework for training protein diffusion models that generates structures with unprecedented diversity and quality. Stateof-the-art generative models are trained on computationally derived structures from AlphaFold2 (AF), as experimentally determined structures are relatively scarce. The resulting models are therefore limited by the quality of synthetic datasets. Since the accuracy of AF predictions degrades with increasing protein length and complexity, de novo generation of long, complex proteins remains challenging. Ambient Protein Diffusion overcomes this problem by treating lowconfidence AF structures as corrupted data. Rather than simply filtering out low-quality AF structures, our method adjusts the diffusion objective for each structure based on its corruption level, allowing the model to learn from both high and low quality structures. Empirically, Ambient Protein Diffusion yields major improvements: on proteins with 700 residues, diversity increases from 45% to 86% from the previous state-of-the-art, and designability improves from 68% to 86%. All of our code, models and datasets are available under the following repository: https://github.com/jozhang97/ambient-proteins.

Recent breakthroughs in machine learning-based structure prediction-most notably Al-phaFold2 [29]-have made it possible to infer accurate protein structures directly from se- * Equal contribution. 39th Conference on Neural Information Processing Systems (NeurIPS 2025).