ZetaDesign: an end-to-end deep learning method for protein sequence design and side-chain packing

Article Multidisciplinary Sciences

Using AlphaFold to predict the impact of single mutations on protein stability and function

Marina A. Pak et al.

Summary: AlphaFold revolutionized the field of structural biology by predicting experiment-quality 3D structures from protein sequences. However, it remains uncertain if AlphaFold can help solve other problems related to protein folding.

PLOS ONE (2023)

Add to Collection

Article Multidisciplinary Sciences

A backbone-centred energy function of neural networks for protein design

Bin Huang et al.

Summary: This study developed a statistical model named SCUBA, which uses neural network-form energy terms to achieve precise design of protein backbones by learning multidimensional, high-order correlations in known protein structures. The results show that using SCUBA for structure design, without using fragments from existing proteins, enables exploration of the designable backbone space and increases the novelty and diversity of de novo designed proteins.

NATURE (2022)

Add to Collection

Article Biochemistry & Molecular Biology

DLPacker: Deep learning for prediction of amino acid side chain conformations in proteins

Mikita Misiura et al.

Summary: The study evaluates the potential of deep neural networks for predicting amino acid side chain conformations, using an image-to-image transformation approach with a U-net style network. The results show that the method outperforms physics-based methods in reconstructing amino acid conformations.

PROTEINS-STRUCTURE FUNCTION AND BIOINFORMATICS (2022)

Add to Collection

Article Multidisciplinary Sciences

Protein sequence design with a learned potential

Namrata Anand et al.

Summary: In this study, a machine learning method and a neural network potential were used to design proteins. The model learned from crystal structure data and was able to automatically generate protein sequences that were experimentally stable. The findings demonstrate the feasibility of using a completely learned approach for protein sequence design.

NATURE COMMUNICATIONS (2022)

Add to Collection

Article Biochemical Research Methods

De novo protein design by an energy function based on series expansion in distance and orientation dependence

Shide Liang et al.

Summary: Despite the challenge in de novo protein design due to the lack of an accurate energy function, an energy function based on series expansions has shown promising results in side-chain and loop prediction accuracy. The method for protein design using this energy function has demonstrated competitive performance compared to existing programs and can effectively recover native residue types and maintain amino acid compositions in designed proteins.

BIOINFORMATICS (2022)

Add to Collection

Article Physics, Multidisciplinary

State-of-the-Art Estimation of Protein Model Accuracy Using AlphaFold

James P. Roney et al.

Summary: Predicting a protein's 3D structure has been a challenge in structural biology, but recent approaches like AlphaFold have made significant progress by combining deep learning and coevolutionary data. We provide evidence that AlphaFold has learned a biophysical energy function and uses coevolution data to solve the global search problem of finding a low-energy conformation. AlphaFold's learned energy function accurately ranks the quality of candidate protein structures without using coevolution data.

PHYSICAL REVIEW LETTERS (2022)

Add to Collection

Article Computer Science, Interdisciplinary Applications

Rotamer-free protein sequence design based on deep learning and self-consistency

Yufeng Liu et al.

Summary: The ABACUS-R method, utilizing an encoder-decoder network, predicts the sidechain type of amino acid sequences from their local environment, simplifying the sequence design process and outperforming energy function-based methods in wet experiments.

NATURE COMPUTATIONAL SCIENCE (2022)

Add to Collection

Article Multidisciplinary Sciences

Protein sequence design by conformational landscape optimization

Christoffer Norn et al.

Summary: The protein design problem aims to find an appropriate amino acid sequence for a desired protein structure, with optimization over all possible sequences and structures using protein structure prediction and backpropagation. The trRosetta model is more effective than Rosetta single-point energy estimations, and combining trRosetta and Rosetta models can result in more funneled energy landscapes.

PROCEEDINGS OF THE NATIONAL ACADEMY OF SCIENCES OF THE UNITED STATES OF AMERICA (2021)

Add to Collection

Article Biochemistry & Molecular Biology

The breakthrough in protein structure prediction

Andrei N. Lupas et al.

Summary: Proteins are essential for all living systems and their structure is crucial for understanding biological processes. The protein folding problem, which has been a challenge for decades, saw a major breakthrough with the use of artificial intelligence technology.

BIOCHEMICAL JOURNAL (2021)

Add to Collection

Article Biochemistry & Molecular Biology

The Stability Landscape of de novo TIM Barrels Explored by a Modular Design Approach

Sergio Romero-Romero et al.

Summary: The study successfully designed a collection of stable de novo TIM barrels (DeNovoTIMs) and observed significant non-additive or epistatic effects when stabilizing mutations from different regions of the barrel were combined. This work is an important step towards the fine-tuned modulation of protein stability by design.

JOURNAL OF MOLECULAR BIOLOGY (2021)

Add to Collection

Article Multidisciplinary Sciences

Highly accurate protein structure prediction with AlphaFold

John Jumper et al.

Summary: Proteins are essential for life, and accurate prediction of their structures is a crucial research problem. Current experimental methods are time-consuming, highlighting the need for accurate computational approaches to address the gap in structural coverage. Despite recent progress, existing methods fall short of atomic accuracy in protein structure prediction.

NATURE (2021)

Add to Collection

Article Biochemical Research Methods