Take a look at the benchmarks today.
New co-folding models come out every week, here's how you keep track of them.
Introducing Tamarind Bio Benchmarks.
Tamarind Benchmarks are a single place to keep track of 10+ of the most popular co-folding tools against a broad set of targets. Here are the highlights:
See leaderboards for a set of benchmarks: Antibody-antigen, protein-ligand, protein-protein complexes, different curated datasets from different authors such as FoldBench and Cofoldarena
Compare the individual biomolecule outputs vs published experimental structures.
Compare two structure prediction methods between each other across benchmarks
Use the Tamarind model router to automatically select the best model for your input.
While our benchmarks show a "best" tool for a given benchmark, we've found that the nuanced differences between the models for a specific input are quite important as well. Last week, we launched our model router, which picks the best tool for your input, not the best tool on average.
With that, we publish the structure prediction results for all PDB entries the models haven't been trained on, meaning you can measure the quality of structures for the closest match to your inputs.
More model types: design + property prediction, new modalities, and more, coming to both Tamarind Benchmarks and Model Router soon!