Case Study: OpenFold advances protein folding research with Weights & Biases

Democratizing AI for Biology With OpenFold

OpenFold, a project dedicated to creating a trainable, open-source implementation of DeepMind's AlphaFold 2, faced the significant challenge of reproducing a complex AI model. With limited initial details available, the OpenFold team needed a highly collaborative tool to manage, log, and track the vast number of experiments required for their investigation. This was critical for their goal of democratizing access to protein folding research.

Weights & Biases provided the solution with its experiment tracking platform, which was integrated via PyTorch Lightning. This gave the OpenFold team full visibility into their machine learning workflows, allowing them to easily superimpose, compare, and debug training runs. The use of Weights & Biases fostered a culture of collaboration and transparency, enabling the team to successfully recreate AlphaFold 2 and share their work with a broad research community, accelerating scientific discovery in biology and beyond.

OpenFold

Gustaf Ahdritz

Lead Developer