Protein folding problem

The protein folding problem is a set of interconnected biophysical and computational challenges traditionally divided into three distinct questions: the folding code, folding kinetics, and structure prediction Verified Answer #3. The folding code refers to Anfinsen's thermodynamic hypothesis regarding how an amino acid sequence determines a final 3D structure Verified Answer #1. Folding kinetics addresses Levinthal's paradox, which questions how a protein chain finds its native conformation rapidly without sampling every possible state Verified Answer #1. Structure prediction involves the computational task of calculating the native 3D atomic coordinates of a protein from its 1D amino acid sequence Verified Answer #3.

Status of Structure Prediction

The computational structure prediction component for stable, globular proteins is widely considered solved Verified Answer #2. AlphaFold 2 achieved a median Global Distance Test (GDT) score of over 90 at the CASP14 competition in 2020, providing accuracy comparable to experimental methods like X-ray crystallography Verified Answer #4. In May 2024, AlphaFold 3 expanded these capabilities by using a diffusion-based architecture to predict holistic biomolecular complexes, including proteins, DNA, RNA, ligands, and ions Verified Answer #1. These AI models succeed by extracting spatial constraints from evolutionary Multiple Sequence Alignments (MSAs) and structural templates found in the Protein Data Bank (PDB) Verified Answer #2.

Unsolved Aspects of Protein Folding

While the thermodynamic endpoint prediction is largely addressed, the physical and chemical principles governing the folding process remain unsolved Verified Answer #1. Current AI tools do not simulate the physical chemistry of atomic interactions over time Verified Answer #2.

Folding Kinetics and Pathways

AI models bypass the physical energy landscape, also known as the folding funnel, to map sequence data directly to a 3D coordinate Verified Answer #1. Consequently, these models cannot generate the kinetic trajectory, which includes intermediate transition states and energy barriers Verified Answer #1. The temporal mechanisms required for a polypeptide chain to collapse into its native state remain unaddressed by static prediction tools Verified Answer #2.

Protein Dynamics and Ensembles

Proteins are dynamic molecules that exist as ensembles of different states rather than static, rigid entities Verified Answer #2. AlphaFold typically produces a single "best" snapshot, often failing to capture the full landscape of functionally relevant alternative conformations Verified Answer #3. This limitation is evident in structural shifts driven by ligand binding; for instance, AlphaFold 3 may predict a closed conformation for certain enzymes regardless of whether a ligand is physically present Verified Answer #2. Because these models rely on static experimental snapshots, they struggle to map structural flexibility and allosteric regulation Verified Answer #2.