Graphormer

Why we included this project

Graphormer is a strong candidate if your work involves predicting molecular or material properties from structure. It wraps a graph-transformer architecture built for molecules into a training toolkit that accepts your own datasets, and it ships pretrained checkpoints trained on large chemical benchmarks like PCQM4M. The code is organized around the same data interfaces used by PyG, DGL, OGB, and the Open Catalyst Project, so researchers already working in those stacks can get started without reworking their pipelines. The same team won the Open Catalyst Challenge, which means the repository shows what actually worked in a competitive materials-science setting, not a toy demo. If you are exploring drug discovery or materials screening, this is a credible, well-tested starting point, though you should plan to bring your own compute and evaluation workflow.