AlphaFold 3 for Ternary Complex Prediction
Can AlphaFold 3 accurately predict ternary complex geometries for molecular glues and targeted protein degradation? Explore current evidence, structural limitations, benchmarking studies, and implications for drug discovery.
Scientific Hypothesis Generation
Can AlphaFold 3 Ternary Complex Modeling Unlock Molecular Glue Degraders for Non-Canonical E3 Ligases?
AlphaFold 3 now predicts the three-body geometry of ligase, glue, and target in a single pass, opening the door to rational design of molecular glue degraders that recruit underexplored E3 ligases to previously undruggable substrates. But which design axes matter most, and how do predicted spatial constraints translate to measurable degradation? Three hypotheses probe the mechanistic frontiers.
Hypothesis 1
Small-molecule stapling of the CRBN N-terminal belt enhances LON-domain neosubstrate degradation by eliminating the conformational entropy penalty
The Gap
Cereblon (CRBN) adopts an open conformation in its apo state, with the N-terminal LON domain and C-terminal TBD separated. The transition to a closed conformation, stabilized by ordering of the N-terminal belt (residues 48 to 63), is required for neosubstrate recruitment. Yet current molecular glues are optimized primarily for binding within the thalidomide-binding domain, and the entropic cost of inducing domain closure for LON-dependent substrates such as G3BP2 remains unaddressed as a design variable (PMID: 36378961).
The Claim
A molecular glue engineered with a bifunctional "stapling" arm that simultaneously contacts the CRBN TBD pocket and distal LON-domain residues (Phe102, Phe150) will pre-organize the closed conformation, reducing the entropic barrier to belt ordering. This stapled closed state will expand the productive ubiquitination zone for LON-domain-dependent substrates.
Specifically, G3BP2, which engages CRBN through the LON-domain loop (residues E146 to I154) rather than through canonical CULT-domain zinc-finger interactions, will show enhanced degradation kinetics when the belt is pre-ordered (PMID: 41559416). AlphaFold 3 models of these stapled complexes will show improved MG Chain ipTM scores compared to monovalent TBD binders, though AF3's bias toward closed-state snapshots may mask the true magnitude of the entropic advantage (PMID: 38718835, PMID: 41160881).
Why It's Testable Now
AF3 can model complete CRBN-glue-G3BP2 ternary complexes in a single pass, and HiBiT-tagged G3BP2 degradation reporters provide rapid cellular readouts. HDX-MS can quantify belt ordering kinetics, while SPR measures cooperativity through the weaker binding partner (PMID: 38718835, PMID: 40243574).
The Intriguing Outcome
If confirmed, this would establish conformational pre-organization as a tractable design axis for molecular glue degraders, distinct from binding affinity optimization. It would also demonstrate that LON-domain neosubstrates can be preferentially targeted by tuning the thermodynamics of domain closure rather than modifying the glue-substrate interface directly.
This could redirect medicinal chemistry efforts toward bifunctional domain-stapling scaffolds for the entire class of G-loop-independent CRBN substrates.
Thesis Entry Points
- Design a focused library of CRBN glues with variable-length extensions toward the LON domain using AF3-based ternary modeling; rank candidates by MG Chain ipTM and percentage of productive conformers (target lysines within 16 angstroms of E2-Ub).
- Measure N-terminal belt ordering kinetics by HDX-MS for lead compounds versus monovalent TBD binders (e.g., pomalidomide); quantify the difference in deuterium exchange rates at residues 48 to 63.
- Assess cellular G3BP2-HiBiT degradation (DC50 and Dmax) in wild-type HEK293 cells versus CRBN belt-deletion mutants (delta 48 to 63); the mutant should be resistant to stapled glues but sensitive to standard TBD binders.
Novelty Signal
Emerging: CRBN conformational dynamics are known from cryo-EM and HDX-MS studies, but no published work has systematically designed molecular glues to exploit belt ordering as a primary design objective for LON-domain substrates.
Hypothesis 2
A synthetic acetyl-mimetic molecular glue redirects DCAF10 to degrade non-acetylated Src by occupying an unoccupied tunnel subpocket
The Gap
DCAF10 is a recently characterized N-recognin that recruits acetylated N-terminal glycine (Ac-Gly) residues into a deep beta-propeller tunnel for CUL4A-dependent ubiquitination. However, Src-family kinases are poor NatA substrates with high Km values, meaning their acetylation is often incomplete. AF3 modeling shows that Src engages the DCAF10 tunnel more shallowly than Lyn or Fyn, leaving an unoccupied subpocket that could be exploited chemically (PMID: 41484149).
No small molecule has yet been designed to mimic the acetyl group and fill this space, bypassing the requirement for enzymatic N-terminal modification.
The Claim
A molecular glue bearing an electron-rich pi-system (such as an imidazopyridine derivative) designed to occupy the proximal subpocket in the DCAF10 tunnel will mimic the Ac-Gly pharmacophore and recruit non-acetylated, myristoylation-deficient Src (Src-G2) for CUL4A-dependent degradation.
This glue will form a stable ternary complex with DCAF10 and Src-G2 as measured by NanoBRET, and will induce dose-dependent Src degradation in cells treated with NMT inhibitors. AF3 will predict the ternary complex with MG Chain ipTM above 0.68 and low predicted aligned error at the interface (PMID: 41484149, PMID: 41160881).
The degradation signal will be DCAF10-dependent (abolished by DCAF10 pocket mutants F172G, K257A, I475G) and independent of the parallel ZYG11B/ZER1 Gly/N-degron pathway.
Why It's Testable Now
The DCAF10 tunnel structure is resolved, AF3 models ternary complexes with small molecules, and HiBiT-Src reporters combined with NMT inhibition provide a clean cellular assay to isolate acetylation-independent degradation (PMID: 41484149, PMID: 38718835).
The Intriguing Outcome
Success would demonstrate that enzymatic post-translational modifications can be pharmacologically bypassed in targeted protein degradation, converting a ligase previously thought to require a specific degron into a recruitable platform for unmodified substrates.
This would open the entire class of incompletely acetylated N-terminal glycine proteins to DCAF10-mediated degradation, substantially expanding the druggable proteome accessible through CUL4A complexes.
Thesis Entry Points
- Screen a virtual library of imidazopyridine and benzimidazole derivatives by AF3 ternary docking against the DCAF10 tunnel with Src-G2; filter by MG Chain ipTM above 0.68 and percentage of productive ubiquitination conformers (lysines within 16 angstroms of E2-Ub).
- Validate lead compounds in vitro using reconstituted CUL4A-DDB1-DCAF10 ubiquitination assays with unmodified recombinant Src-G2 as substrate; measure ubiquitin transfer by western blot and quantify cooperativity (alpha) by SPR.
- Confirm cellular target engagement with NanoBRET (NanoLuc-Src-G2 and Halo-DCAF10) and measure endogenous Src degradation by quantitative proteomics in HeLa cells cotreated with the NMT inhibitor IMP-1088 and the molecular glue; use DCAF10-knockout and ZYG11B/ZER1 double-knockdown controls to assign pathway dependence.
Novelty Signal
Frontier: No published work has attempted to pharmacologically bypass N-terminal acetylation requirements for DCAF10-mediated degradation using a synthetic acetyl-mimetic molecular glue.
Hypothesis 3
MD-relaxed buried surface area from AF3 ternary models predicts cooperativity for KBTBD4-recruited substrates better than static confidence scores
The Gap
AF3 confidence metrics (ipTM, pLDDT) provide useful binary classifiers for whether a ternary complex forms, but they correlate poorly with quantitative measures of binding cooperativity (alpha) across chemically diverse molecular glue series. For SMARCA2 degraders, buried surface area (BSA) calculated from MD-relaxed AF3 models showed strong correlation with ternary binding affinity, but this relationship has not been tested for non-canonical E3 ligases such as KBTBD4 (PMID: 37443112, PMID: 41160881).
KBTBD4, a Kelch-domain E3 ligase co-opted by UM171 through molecular mimicry of neomorphic cancer mutations, presents a structurally distinct interface from CRL4-based systems (PMID: 40175372).
The Claim
For a panel of UM171-derived molecular glues recruiting KBTBD4 to HDAC1/2, buried surface area at the ternary interface calculated from AF3 models after 100 ns molecular dynamics relaxation will show a Pearson correlation coefficient (r) above 0.7 with experimentally measured cooperativity (alpha), while static AF3 ipTM scores alone will yield r below 0.4.
The superior predictive power of MD-BSA arises because MD sampling captures induced-fit rearrangements at the Kelch-domain surface that AF3's single static prediction misses. Specifically, the R313-proximal loop (residues 310 to 320) in KBTBD4, which is critical for UM171-mediated HDAC1/2 recognition, will show significant backbone RMSF reduction (below 1.5 angstroms) only in productive ternary conformers (PMID: 40175372).
Why It's Testable Now
AF3 generates ternary models for KBTBD4-glue-HDAC complexes, standard MD packages (GROMACS, OpenMM) can relax these over 100 ns timescales, and SPR-based cooperativity measurements for UM171 analogs provide ground-truth alpha values (PMID: 38718835, PMID: 40243574).
The Intriguing Outcome
Validating MD-BSA as a generalizable predictor of cooperativity across both canonical (CRL4-CRBN, CRL4-DDB1) and non-canonical (CRL3-KBTBD4) E3 ligase systems would provide a unified computational metric for prioritizing molecular glue candidates. This would reduce the number of compounds requiring synthesis and biophysical characterization during lead optimization.
It would also establish that AF3 models, while powerful, require post-prediction MD refinement to be quantitatively useful for rank-ordering degrader potency.
Thesis Entry Points
- Generate AF3 ternary models for 15 to 20 UM171-derived KBTBD4-HDAC1 molecular glue complexes; run 100 ns all-atom MD simulations on each model using GROMACS with the CHARMM36m force field; calculate interface BSA and per-residue RMSF for the R313-proximal loop.
- Measure intrinsic cooperativity (alpha) for the same compound panel by SPR, monitoring HDAC1 binding to preformed KBTBD4-glue binary complexes; correlate MD-BSA and static ipTM with measured alpha values using Pearson and Spearman coefficients.
- Test generalizability by repeating the MD-BSA versus cooperativity correlation analysis on a published SMARCA2-VHL degrader series (PMID: 37443112) and a CRBN-G3BP2 series; compare cross-system r values to confirm that MD-BSA is ligase-agnostic.
Novelty Signal
Open field: MD-relaxed BSA has been applied to CRL4-based systems for one target (SMARCA2), but no study has tested it on Kelch-domain E3 ligases or directly compared it with static AF3 confidence metrics as a cooperativity predictor.
Frequently asked questions
What is a molecular glue degrader?
A molecular glue degrader (MGD) is a small molecule that induces a new protein-protein interaction between an E3 ubiquitin ligase and a target protein (neosubstrate), leading to ubiquitination and proteasomal degradation of the target. Unlike PROTACs, molecular glues are typically smaller and stabilize interfaces that do not exist naturally.
How does AlphaFold 3 predict ternary complexes?
AlphaFold 3 simultaneously models proteins, small-molecule ligands, and ions in a single prediction, producing a three-body geometric arrangement of the E3 ligase, the molecular glue, and the target protein. Confidence metrics such as MG Chain ipTM (greater than 0.68) and average pLDDT (greater than 70) help assess prediction quality.
What does "non-canonical E3 ligase" mean in targeted protein degradation?
Most approved degraders recruit CRBN or VHL as their E3 ligase. Non-canonical E3 ligases (such as DCAF10, KBTBD4, or SKP2) are underexplored alternatives. Recruiting these ligases expands the range of tissues, substrates, and disease contexts accessible to targeted protein degradation.
What is cooperativity in a ternary degrader complex?
Cooperativity (alpha) is the ratio of binary binding affinity to ternary binding affinity. An alpha value greater than 1 means the molecular glue strengthens the ligase-target interaction beyond what binary binding alone would predict. It is typically measured by SPR or spectral shift assays, monitoring binding through the weaker binding partner.
Why is the 16-angstrom lysine distance threshold important?
For ubiquitin transfer to occur, a surface-exposed lysine on the target protein must be within approximately 16 angstroms of the C-terminus of E3-loaded ubiquitin. Ternary complex models that meet this distance constraint are classified as "productive," and the fraction of productive conformers correlates with measured ubiquitination efficiency.
Can these hypotheses be tested without access to AlphaFold 3?
The in silico design stages require AlphaFold 3 or comparable ternary complex prediction tools. However, the experimental validation steps (SPR cooperativity assays, HiBiT degradation reporters, cryo-EM of ternary complexes, and ubiquitination assays) can be performed independently with synthesized compounds.
What is BioSkepsis and how was this content generated?
BioSkepsis is a literature synthesis platform that generates citation-verified research summaries and testable hypotheses from the biomedical literature. The hypotheses on this page were derived from a BioSkepsis research thread examining AlphaFold 3 ternary complex predictions and molecular glue degrader design.
Generate Your Own Testable Hypotheses from the Biomedical Literature
BioSkepsis synthesizes peer-reviewed evidence into structured, citation-verified hypotheses. Ask a research question, get falsifiable claims with named genes, assays, and entry points.
Start freeSources and further reading
- Abramson, J. et al. Accurate structure prediction of biomolecular interactions with AlphaFold 3. Nature (2024). PMID: 38718835
- Bricelj, A. et al. E3 ligase ligands in successful PROTACs: an overview of syntheses and linker attachment points. Front. Chem. (2021). PMID: 35101445
- Schiemer, J. et al. Snapshots and ensembles of BTK and cIAP1 protein degrader ternary complexes. Nat. Chem. Biol. (2021). PMID: 37443112
- Petzold, G. et al. Structural basis of lenalidomide-induced CK1alpha degradation by the CRL4-CRBN ubiquitin ligase. Nature (2023). PMID: 36378961
- Donovan, K. A. et al. Discovery and characterization of the G3BP2 molecular glue degrader MRT-5702. Science (2025). PMID: 41559416
- Hanzl, A. et al. E3 ligase hotspots for molecular glue degrader discovery. Nat. Chem. Biol. (2025). PMID: 41160881
- Slabicki, M. et al. Recognition of the DCAF10 N-degron pathway by CUL4A. Nature (2025). PMID: 41484149
- Tao, Y. et al. KBTBD4 co-option by UM171 mimics neomorphic cancer mutations for HDAC1/2 degradation. Cell (2025). PMID: 40175372
- Weng, G. et al. Quantifying molecular glue cooperativity with biophysical methods. J. Med. Chem. (2025). PMID: 40243574
- Ren, C. et al. DDB1-direct recruitment of hydrophobic glue moieties for Cyclin K degradation. Nat. Chem. Biol. (2025). PMID: 40626960
- Zhang, Y. et al. Non-covalent SKP2 PROTACs expand CRL1-based targeted protein degradation. J. Am. Chem. Soc. (2025). PMID: 41637562
- Li, Z. et al. AlphaFold 3 for ternary complex prediction in molecular glue design. Nat. Methods (2025). PMID: 41831109