High-Performance Deep Learning Toolbox for Genome-Scale Prediction of Protein Structure and Function

Gao, Mu; Lund-Andersen, Peik K.; Morehead, Alex; Mahmud, Sajid; Chen, Chen; Xiao, Chen; Giri, Nabin; Roy, Raj; Quadir, Farhan; Effler, T. Chad; Prout, Ryan; Abraham, Subil; Elwasif, Wael; Haas, Nicholas Quentin; Skolnick, Jeffrey; Cheng, Jianlin; Sedova, Ada

doi:10.1109/mlhpc54614.2021.00010

Cited by 11 publications

(7 citation statements)

References 70 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Multimer-AF2 dataset. Drawing loose inspiration from [58], for this work, we assembled the new Multimer-AF2 (MAF2) dataset comprised of multimeric structures predicted by AlphaFold 2 [8] and AlphaFold-Multimer [7], using the latest structure prediction pipeline for the Summit supercomputer [59,60]. Originating from the EVCoupling [50] and DeepHomo [51] datasets, the proteins for which we predicted structures using a combination of both AlphaFold methods consist of heteromers and homomers, respectively.…”

Section: A Additional Resultsmentioning

confidence: 99%

A Gated Graph Transformer for Protein Complex Structure Quality Assessment and its Performance in CASP15

Chen

Morehead

Liu

et al. 2022

Preprint

Self Cite

View full text Add to dashboard Cite

Proteins interact to form complexes to carry out essential biological functions. Computational methods have been developed to predict the structures of protein complexes. However, an important challenge in protein complex structure prediction is to estimate the quality of predicted protein complex structures without any knowledge of the corresponding native structures. Such estimations can then be used to select high-quality predicted complex structures to facilitate biomedical research such as protein function analysis and drug discovery. We challenge this significant task with DProQ, which introduces a gated neighborhood-modulating Graph Transformer (GGT) designed to predict the quality of 3D protein complex structures. Notably, we incorporate node and edge gates within a novel Graph Transformer framework to control information flow during graph message passing. We train and evaluate DProQ on four newly-developed datasets that we make publicly available in this work. Our rigorous experiments demonstrate that DProQ achieves state-of-the-art performance in ranking protein complex structures.3

show abstract

Section: A Additional Resultsmentioning

confidence: 99%

A Gated Graph Transformer for Protein Complex Structure Quality Assessment and its Performance in CASP15

Chen

Morehead

Liu

et al. 2022

Preprint

Self Cite

View full text Add to dashboard Cite

show abstract

“…We use the GPU computing resource on the Summit supercomputer provided by Oak Ridge National Laboratory (ORNL) to train the deep learning network above. The Summit cluster [40, 41] provides many compute nodes each having 6 GPUs and 16 GB of memory, which enables the distributed deep learning training. We train 2D U-Net, 1D U-Net, and the multi-head attention layer on three separate GPU nodes.…”

Section: Methodsmentioning

confidence: 99%

Multi-Head Attention-based U-Nets for Predicting Protein Domain Boundaries Using 1D Sequence Features and 2D Distance Maps

Mahmud

Guo

Quadir

et al. 2022

Preprint

Self Cite

View full text Add to dashboard Cite

The information about the domain architecture of proteins is useful for studying protein structure and function. However, accurate prediction of protein domain boundaries (i.e., sequence regions separating two domains) from sequence remains a significant challenge. In this work, we develop a deep learning method based on multi-head U-Nets (called DistDom) to predict protein domain boundaries utilizing 1D sequence features and predicted 2D inter-residue distance map as input. The 1D features contain the evolutionary and physicochemical information of protein sequences, whereas the 2D distance map includes the structural information of proteins that was rarely used in domain boundary prediction before. The 1D and 2D features are processed by the 1D and 2D U-Nets respectively to generate hidden features. The hidden features are then used by the multi-head attention to predict the probability of each residue of a protein being in a domain boundary, leveraging both local and global information in the features. The residue-level domain boundary predictions can be used to classify proteins as single-domain or multi-domain proteins. It classifies the CASP14 single-domain and multi-domain targets at the accuracy of 72.7%, 8.02% more accurate than the state-of-the-art method. Tested on the CASP14 multi-domain protein targets with expert annotated domain boundaries, the average per-target F1 measure score of the domain boundary prediction by DistDom is 0.241, 18.72% higher than the state-of-the-art method.

show abstract

“…We employed the ESM-2 pretrained model, which comprises 33 layers and 650M parameters, to generate embedding dimensions of 1,280. This computation was performed on the CPU of the Andes supercomputer 39 , utilizing its 256 GB RAM. To accommodate memory constraints, we utilized a sequence cutoff of 2,750 residues for each protein sequence.…”

Section: Training and Testing Deep Transformer Models On Cryo2structdatamentioning

confidence: 99%

Cryo2StructData: A Large Labeled Cryo-EM Density Map Dataset for AI-based Modeling of Protein Structures

Giri

Cheng

2023

Preprint

Self Cite

View full text Add to dashboard Cite

The advent of single-particle cryogenic electron microscopy (cryo-EM) has brought forth a new era of structural biology, enabling the routine determination of large biological protein complexes and assemblies at atomic resolution. The high-resolution structures of protein complexes and assemblies significantly expedite biomedical research and drug discovery. However, automatically and accurately reconstructing protein structures from high-resolution density maps generated by cryo-EM is still time-consuming and challenging when template structures for the protein chains in a target protein complex are unavailable. Artificial intelligence (AI) methods such as deep learning trained on limited amounts of labeled cryo-EM density maps generate unstable reconstructions. To address this issue, we created a dataset called Cryo2Struct consisting of 7,600 preprocessed cryo-EM density maps whose voxels are labelled according to their corresponding known protein structures for training and testing AI methods to infer protein structures from density maps. It is larger and has better quality than any existing, publicly available dataset. We trained and tested deep learning models on Cryo2Struct to make sure it is ready for the large-scale development of AI methods for reconstructing protein structures from cryo-EM density maps. The source code, data and instructions to reproduce our results are freely available at https://github.com/BioinfoMachineLearning/cryo2struct.

show abstract

High-Performance Deep Learning Toolbox for Genome-Scale Prediction of Protein Structure and Function

Cited by 11 publications

References 70 publications

A Gated Graph Transformer for Protein Complex Structure Quality Assessment and its Performance in CASP15

A Gated Graph Transformer for Protein Complex Structure Quality Assessment and its Performance in CASP15

Multi-Head Attention-based U-Nets for Predicting Protein Domain Boundaries Using 1D Sequence Features and 2D Distance Maps

Cryo2StructData: A Large Labeled Cryo-EM Density Map Dataset for AI-based Modeling of Protein Structures

Contact Info

Product

Resources

About