Protein Dynamics Beyond Structure Prediction
1 The Roadmap argues that AlphaFold-level structure prediction solved “where proteins end up,” but not “how they get there”: folding pathways, kinetics, intermediates, assembly routes, and misfolding competition remain largely non-predictive from sequence alone.
2 Central thesis: proteins are dynamic, stochastic ensembles shaped by context (translation, chaperones, crowding, membranes, PTMs, stress), so a single static structure is an incomplete target; the real target is an evolving conformational landscape plus transition kinetics.
3 It frames a “structure prediction paradox”: models can often predict one (or a few) stable states, yet cannot predict (i) the full ensemble, (ii) barrier heights and rate constants, (iii) pathway heterogeneity, (iv) aggregation propensity, or (v) cellular outcomes of kinetic partitioning.
4 A key emphasis is multiscale coupling: Å-level interactions and ns–µs fluctuations connect to ms–s folding transitions, while misfolding/aggregation can accumulate over years; bridging these time/length scales is positioned as the core methodological gap.
5 The article highlights why in vivo folding is fundamentally different from typical in vitro refolding: co-translational, vectorial emergence from the ribosome; codon-dependent translation speeds; ribosome exit-tunnel constraints; and early engagement by ribosome-associated factors and chaperone systems that reshape pathways rather than merely stabilizing endpoints.
6 Misfolding is treated as a kinetic competition problem, not a “wrong structure” problem: the native fold can be correct yet still lose to off-pathway aggregation under stress/aging or limited proteostasis capacity; amyloid polymorphism (many distinct fibril folds from the same sequence) is presented as a major unpredictability from static models.
7 Translational motivation: for Alzheimer’s, ALS, systemic amyloidoses, CFTR misfolding, and proinsulin variants, detecting aggregated proteins or biomarkers is often insufficient to predict trajectory; the Roadmap argues for mechanistically grounded markers tied to specific conformers and kinetic vulnerabilities.
8 Experimental direction: the field is moving from ensemble averages (CD, FTIR, stopped-flow, NMR, HDX-MS, SAXS/cryo-EM) toward systematic single-molecule trajectory measurements (single-molecule fluorescence/FRET; force spectroscopy such as optical tweezers/AFM) to directly observe intermediates, heterogeneity, and barrier crossings.
9 Computational direction: atomistic MD offers mechanism but is timescale-limited; coarse-grained/multiscale models extend reach but need validation; machine learning is positioned as a bridge to integrate heterogeneous data and accelerate sampling—yet predicting dynamics is described as a fundamentally different challenge than predicting structures.
10 Proposed ecosystem shift: replicate what enabled structure prediction (shared standards, curated datasets, benchmarking) but for folding trajectories and kinetics; the Roadmap advocates iterative experiment–model feedback loops to connect sequence to pathways, ensembles, assembly/misfolding, and ultimately cellular phenotypes.
📜Paper:
arxiv.org/abs/2606.08647
#ProteinFolding #ProteinDynamics #SingleMolecule #MolecularDynamics #MachineLearning #Proteostasis #Chaperones #Misfolding #Amyloid #ComputationalBiology