PathFold is a generative AI framework that learns and predicts how proteins fold and misfold over time, directly from their amino acid sequence.
Beyond static structures
Protein misfolding, not structural damage per se, is the kinetic failure behind Alzheimer's, Parkinson's and related proteinopathies. Early off-pathway conformations of amyloid-β, tau and α-synuclein precede clinical symptoms by years, yet no existing computational tool models this process at scale.
Single end-point structure
State-of-the-art structure predictors describe only the folding endpoint: one lowest-energy conformation, without kinetic or temporal information, and they capture fewer than one in seven experimentally validated fold-switching conformations.
Too short timescales
MD simulations generate atomistic folding trajectories, but clinically relevant folding and misfolding events happen on millisecond-to-second timescales that remain out of reach even with enhanced sampling and Markov state models.
Kinetically ordered pathways from sequence
PathFold models folding as a coarse-grained, kinetically ordered sequence of structural transitions, capturing the order, timing and interdependence of conformational events that decide between a stable native state and misfolding.
Interactive
Pick a benchmark protein or paste your own sequence to see a schematic of how PathFold reasons: residue-level propensity tracks, a kinetically ordered trajectory and the point where productive folding and off-pathway misfolding diverge.
10-residue designed β-hairpin; Tier-1 fast-folder benchmark · 10 residues
The framework
PathFold combines the representational power of protein language models with constraints from energy-landscape theory, and ships everything as open, reusable artefacts.
Trajectory generation model
A transformer-based generative model that predicts coarse-grained folding trajectories from amino acid sequences while preserving kinetic order, intermediate-state plausibility and thermodynamic consistency along the whole pathway.
Learn moreMisfolding & aggregation model
A probabilistic model that treats misfolding as a divergence from normal folding trajectories and predicts residue-resolved misfolding and aggregation risk with calibrated uncertainty.
Learn moreOpen platform & benchmark
A modular open-source software suite that integrates both models, plus DynBench, the first standardised benchmarking framework with a public leaderboard for sequence-to-trajectory models.
Learn moreLatest
Project updates will be published here.
Selected
Papers, preprints and datasets will appear here.
Consortium
PathFold is coordinated at Saarland University and supported by two associated partners that anchor the project in Germany's AI and neurodegenerative-disease research landscape.
PathFold is funded by the German Federal Ministry of Research, Technology and Space (BMFTR) under the “Zukunft eHealth” funding guideline (grant no. 031L0342).