DiAMoNDBack: Diffusion-Denoising Autoregressive Model for Non-Deterministic Backmapping of Cα Protein Traces

Abstract

Coarse-grained molecular models of proteins permit access to length and time scales unattainable by all-atom models and the simulation of processes that occur on long time scales, such as aggregation and folding. The reduced resolution realizes computational accelerations, but an atomistic representation can be vital for a complete understanding of mechanistic details. Backmapping is the process of restoring all-atom resolution to coarse-grained molecular models. In this work, we report DiAMoNDBack (Diffusion-denoising Autoregressive Model for Non-Deterministic Backmapping) as an autoregressive denoising diffusion probability model to restore all-atom details to coarsegrained protein representations retaining only Cα coordinates. The autoregressive generation process proceeds from the protein Nterminus to C-terminus in a residue-by-residue fashion conditioned on the Cα trace and previously backmapped backbone and sidechain atoms within the local neighborhood. The local and autoregressive nature of our model makes it transferable between proteins. The stochastic nature of the denoising diffusion process means that the model generates a realistic ensemble of backbone and sidechain all-atom configurations consistent with the coarse-grained Cα trace. We train DiAMoNDBack over 65k+ structures from the Protein Data Bank (PDB) and validate it in applications to a hold-out PDB test set, intrinsically disordered protein structures from the Protein Ensemble Database (PED), molecular dynamics simulations of fast-folding mini-proteins from DE Shaw Research, and coarse-grained simulation data. We achieve state-of-the-art reconstruction performance in terms of correct bond formation, avoidance of side-chain clashes, and the diversity of the generated side-chain configurational states. We make the DiAMoNDBack model publicly available as a free and open-source Python package.

Application of Generative Artificial Intelligence in Predicting Membrane Partitioning of Drugs: Combining Denoising Diffusion Probabilistic Models and MD Simulations Reduces the Computational Cost to One-Third

“…By comparing the quality of the generated data with original data, DDPMs were accurate in most cases up to using only 10 out of 31 windows and observing sustained accuracy while cutting the data down to one-third of what is traditionally required points to the strong ability of DDPMs to learn the underlying data distribution. This observation concurs with the strong performance of DDPMs in various tasks in various fields, including replica-exchange simulations, docking, protein structure prediction, ,, and other related tasks. , − Furthermore, in addition to the comparison with the ground truth established by all 31 window US simulations, we compared the performance of the DDPM-enhanced model with only 10 US data. As expected, the PMF curve generated from only 10 US data was rough and produced large error values for both Δ G -partitioning and Δ G -crossing (Figure S6).…”

Data-Efficient Generation of Protein Conformational Ensembles with Backbone-to-Side-Chain Transformers

“…For example, Stieffenhofer et al trained generative adversarial networks to reconstruct protein structural ensembles. Similar tasks have been pursued with different architectures. , Our previous work seeks not only to add stochasticity to this task but also to ensure that the sampled distribution can be reweighted to ensure Boltzmann statistics; we carry this out by training normalizing flows that conditionally sample rotamers given a backbone conformation . However, unlike refs and , the models developed in ref trained for individual proteins and are not general purpose.…”

“…Due to the tractable structure of rotamer distributions, several groups have pursued efforts to systematically enumerate the available states. , Furthermore, these libraries have subsequently been deployed to capture the sequence dependence of side chains using relaxation algorithms. ,, Recently, this topic has been revisited in the context of more expressive probabilistic models built with deep neural networks. ,,, However, in large part, these efforts have not sought to add physically meaningful thermal stochastic fluctuations.…”

MinActionPath2: path generation between different conformations of large macromolecular assemblies by action minimization

“…In practice, a full atom description of large macromolecular nanomachines is currently difficult to achieve and one must resort to coarse-grained models ( 5–7 ). Note that several methods have been developed to go from a coarse-grained representation of biological macromolecules back to their full atomic models ( 8–10 ).…”