Cheminformatics

FoldKit: A Python library for efficient storage and retrieval of co-folding predictions

Jonathan A. Levine, Melissa Pathil, Samuel Nitz, Olga Lyudovyk, Benjamin D. Greenbaum

Featured September 7, 2026

AI-generated analysis — This is SciGrove's AI interpretation of the paper, not peer-reviewed content. Always refer to the original paper.

Simply

FoldKit helps scientists manage huge AlphaFold 3 protein prediction files by squishing them into tiny, organized packages, making it much easier to store and quickly check how well proteins are predicted to stick together.

In depth

The paper introduces FoldKit, a Python library designed to address the challenges of managing large datasets generated by AlphaFold 3 (AF3) co-folding predictions. It converts raw AF3 outputs, which are often inefficient JSON files, into a compact, structured representation using NPZ files. This significantly reduces storage requirements, by 5-15 fold, while providing programmatic access to various confidence metrics (e.g., pLDDT, pTM, ipTM, ipAE, ipSAE) for individual predictions and ensembles.

Key Takeaways

1
FoldKit reduces storage requirements for AlphaFold 3 co-folding outputs by 5-15 fold by converting raw JSON into a compact, structured NPZ format.
2
The library provides a programmatic API for easy access and aggregation of global, single chain, and interchain confidence metrics (e.g., pLDDT, ipTM, ipAE).
3
FoldKit's storage savings are particularly significant for large multi-chain complexes and pooled co-folding approaches, where pairwise confidence matrices dominate raw output size.

Conceptual Flow

HIGH LEVEL

Converting Bulky Files to Compact Data

The tool takes big, messy prediction files and turns them into small, organized data packages that are easy to use.

Big Prediction Files

Shrink and Organize

Small Data Packages

Saving Space and Speeding Up Research

By making files smaller, the tool saves a lot of computer space and helps scientists study many more protein interactions faster.

Lots of Storage Used

Slow Data Access

Reduce and Speed Up

Much Less Storage

Fast Data Access

This breakdown was generated by SciGrove. Get the same analysis — intuition, storyboard, peer review, a runnable prototype and a glossary — on any paper you upload or paste a DOI for.