PRosettaC outperforms AlphaFold3 for modeling PROTAC ternary complexes

Abstract

Targeted protein degradation via PROTACs offers a promising therapeutic strategy, yet accurate modeling of ternary complexes remains a critical challenge in degrader design. In this study, we systematically benchmark two leading structure prediction tools, AlphaFold-3 and PRosettaC, against a curated dataset of 36 crystallographically resolved ternary complexes. Using DockQ as a quantitative interface scoring metric, we assess the structural fidelity of predicted complexes under both scaffold-inclusive and stripped configurations. Our results demonstrate that AlphaFold-3's performance is often inflated by accessory proteins such as Elongin B/C or DDB1, which contribute to overall interface area but not degrader-specific binding. PRosettaC, on the other hand, leverages chemically defined anchor points to yield more geometrically accurate models in select systems, though it frequently fails when linker sampling is insufficient or misaligned. To overcome the limitations of static benchmarking, we introduce a dynamic evaluation strategy using molecular dynamics simulations of the crystal structures. This frame-resolved analysis reveals that several PRosettaC models, while poorly aligned to the static crystal conformation, transiently achieve high DockQ alignment with specific frames along the MD trajectory. These findings underscore the importance of incorporating protein flexibility into benchmarking workflows and suggest that transient conformational compatibility may be overlooked in conventional evaluations. By combining constraint-based modeling with dynamic frame matching, this study provides a more nuanced framework for assessing ternary complex predictions and informs the selection of in silico tools for rational PROTAC development.

Keywords: AlphaFold; Computational modeling; PROTACs; PRosettaC; Protein modeling; Structure-based-drug-design.

Fig. 1

Median-centered DockQ score distributions per method for each reference benchmark system. PRosettaC displayed broader score variability and higher medians in several cases, especially 6BN7, 6ZHC, 7KHH, and 8QVU. AF3 Core and Minimum predictions clustered tightly at relatively low DockQ scores.

Fig. 2

Cumulative distribution function (CDF) of DockQ scores across all predictions. AF3 Minimal and Core models exhibited highly skewed distributions toward low DockQ values, while PRosettaC maintained a longer high-score tail suggesting better predictions in many cases.

Fig. 3

Normalized distribution of DockQ scores across the three prediction methods: AF3 Minimal and Core Complex models, and PRosettaC (Rosetta-based docking) Ternary Complex predictions. Each bar represents the fraction of models within a given DockQ bin, normalized within each group. The majority of models across all methods fell below the 0.1 DockQ threshold. PRosettaC showed a broader distribution and slightly higher frequencies in mid-to-high score bins, reflecting improved modeling of native-like interfaces in certain systems.

Fig. 4

Heat map of median DockQ score differences. Rows list each head-to-head comparison in the format “Method A—Method B,” while columns show the benchmark complexes (PDB IDs). Cell colors represent the median DockQ of Method A minus Method B: red indicates Method A performed better, blue indicates Method B performed better, and grey denotes negligible difference. The plot highlights that PRosettaC Ternary predictions in most cases exceeded AF3 Minimal results, whereas AF3 Core models rarely outperformed PRosettaC.

Fig. 5

Swarm plot of DockQ scores per model. PRosettaC consistently populated the high-scoring regime (> 0.5), particularly for 6BN7, 6ZHC, and 7KHH, while AF3 models clustered near baseline.

Fig. 6

Median DockQ score per method for each native complex. PRosettaC outperforms in nearly half of the targets, with some competition from AF3 Core Complex (e.g., 6KHH, 5T35) and Ternary (e.g., 7Z77, 6HAX).

Fig. 7

Frame-by-frame DockQ scores for 6HAX (SMARCA2:VHL) over a 50 ns MD trajectory (5000 frames) as an illustrative example. Each colored trace reports the similarity between a static AF3 Minimal Complex model and successive MD frames of the crystal structure. Although Model #1 exhibited a brief peak at DockQ = 0.10 early in the simulation, the native complex did not sample conformations close to any AF3 model. All traces fluctuated below 0.1 throughout, indicating that crystal-derived dynamics did not traverse the predicted interfaces.

Fig. 8

Frame-by-frame DockQ scores for the five best PRosettaC models evaluated against a 50 ns MD ensemble of the SMARCA2:VHL ternary complex (PDB 6HAX). The reference trajectory was sampled every 50 frames, and each colored trace shows the DockQ score of a single PRosettaC model against the MD time course of the PDB reference ternary complex. Although most frames produced modest agreement, several stretches exceeded the 0.23 “acceptable” threshold, peaking at 0.45, revealing transient conformations in the native ensemble that align well with the PRosettaC predictions.