Use Binding ddG Online
Commercially Available Binding ddG No-Code Web Server
Binding ddG: Predicting Mutation Effects on Protein-Protein Binding Affinity with the Rotamer Density Estimator
Binding ddG on Tamarind Bio is based on the Rotamer Density Estimator (RDE), a conditional normalizing flow that models side-chain conformations to predict how mutations change protein-protein binding free energy (ddG). The work, titled "Rotamer Density Estimator is an Unsupervised Learner of the Effect of Mutations on Protein-Protein Interaction", is by Shitong Luo, Yufeng Su, Zuofan Wu, Chenpeng Su, Jian Peng and Jianzhu Ma, and appeared at ICLR 2023 with a bioRxiv preprint. The official code is released under the Apache-2.0 license.
Measuring the effect of every possible mutation on a protein complex is slow and expensive, and only a few thousand annotated mutations are publicly available, which limits purely supervised learning. Sequence-based methods also struggle for general protein-protein interactions because of limited evolutionary information. RDE instead uses structure and a physical idea: binding usually reduces side-chain conformational entropy at the interface, and comparing that entropy loss between wild-type and mutant complexes estimates ddG.
Property | Detail |
|---|---|
Core model | Rotamer Density Estimator (conditional normalizing flow on side-chain torsions) |
Predictors | RDE-Linear (entropy-based) and RDE-Network (neural network on RDE representations) |
Input | Wild-type complex structure and the mutations to evaluate |
Output | Predicted ddG values and per-residue conformational entropy estimates |
Training data | PDB-REDO for RDE; SKEMPI2 for ddG |
How the Rotamer Density Estimator Predicts Binding Free Energy Changes
Learn rotamer densities: an encoder with invariant self-attention produces a representation per residue, and spline-based flows model the density of side-chain torsion angles.
Estimate entropy: rotamers are sampled from the flow, and the average negative log-probability gives conformational entropy.
RDE-Linear: estimates entropy loss on binding for wild-type and mutant states with learned coefficients and per-amino-acid biases.
RDE-Network: takes frozen RDE representations, subtracts wild-type from mutant, and an MLP predicts ddG; anti-symmetry is enforced by averaging with the swapped prediction.
The mutant structure is not required as input, which makes it practical to scan many candidate mutations.
Binding ddG Benchmarks and Case Study on SARS-CoV-2 Antibody Mutations
On the SKEMPI2 benchmark, the authors report that RDE-Network outperforms all baselines, with notable gains in per-structure correlations, while RDE-Linear is comparable to Rosetta. In a case study on an anti-SARS-CoV-2 antibody with 494 possible single-point mutations, five are known to be beneficial; RDE-Network and DDGPred each rank three of them in the top 10%, with RDE-Network ranking them higher. Estimated entropy also correlates with side-chain B-factors (average Pearson 0.4637, Spearman 0.4282), and the density estimator beats Rosetta fixbb and SCWRL4 on side-chain torsion angle prediction.
What is Tamarind Bio?
Tamarind Bio is a no-code bioinformatics platform built to give life scientists and researchers access to powerful computational tools. Many cutting-edge machine learning models are hard to deploy and use. Tamarind provides an intuitive, web-based environment that removes the complexity of high-performance computing, software dependencies and command-line interfaces.
The platform is designed for biologists, chemists and other researchers who may not have a background in programming or cloud infrastructure but want to run models on their own data. Key features include:
A user-friendly graphical interface for setting up and launching experiments
A robust API for integration into existing research pipelines
An automated system for managing and scaling computational resources
Tamarind treats information and data security as a top priority, as detailed in its Trust Center and Terms of Service.
Accelerating Discovery with Binding ddG on Tamarind Bio
Antibody optimization: rank point mutations in an antibody-antigen interface for improved binding.
Interface mutation scanning: prioritize substitutions that strengthen or weaken a protein-protein interaction.
Entropy insights: inspect per-residue conformational entropy to understand how binding restricts side chains.
Rotamer sampling: use the density estimator to predict side-chain conformations.
How to Use Binding ddG on Tamarind Bio
Open the tool: log in to tamarind.bio and open Binding ddG.
Upload the complex: provide the wild-type protein-protein complex structure.
Define the partners: specify which chains form each side of the interaction, if exposed.
Specify mutations: list the substitutions to evaluate by chain, position and residue.
Choose a predictor: select RDE-Network or RDE-Linear, if exposed.
Run the job: submit and wait for predictions.
Download and interpret: review predicted ddG for each mutation to rank candidates.
Use downstream: validate top-ranked mutations experimentally or with further modeling.
Things to Keep in Mind
The model does not directly model backbone flexibility.
RDE-Linear does not capture the nonlinear effects that seem to dominate multi-point mutations, so it underperforms on the full dataset.
The approach relies on structures rather than sequence alone.
Entropy estimates correlate only moderately with B-factors.
Source: Luo, S., Su, Y., Wu, Z., Su, C., Peng, J. and Ma, J., Rotamer Density Estimator is an Unsupervised Learner of the Effect of Mutations on Protein-Protein Interaction. ICLR 2023; bioRxiv preprint DOI: 10.1101/2023.02.28.530137. Code: https://github.com/luost26/RDE-PPI