Accurate Protein Structure Prediction by Embeddings and Deep Learning Representations

Drori, Iddo; Thaker, Darshan D.; Srivatsa, Arjun; Jeong, Daniel; Wang, Yueqi; Nan, Linyong; Wang, Fan; Leggas, Dimitri; Lei, Jinhao; Lu, Weiyi; Fu, Weilong; Gao, Yuan; Karri, Sashank; Kannan, A.; Moretti, Antonio; AlQuraishi, Mohammed; Keasar, Chen; Pe’er, Dana

doi:10.48550/arxiv.1911.05531

Cited by 1 publication

(1 citation statement)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Wang et al [27] proposed Secondary Structure Recurrent Encoder-Decoder Networks to solve the problem of protein secondary structure prediction. Iddo et al [28] demonstrated the structure prediction results of the distance matrix and the torsion angle predicted using the deep learning model, which replaced the results of the winning team in CASP12. Senior et al [29] proposed a new challenge system-AlphaFold-which has the largest number of correctly predicted structures in the free modeling (FM) class of CASP13, representing a major advancement in protein structure prediction.…”

Section: Protein Prediction-based Deep Learningmentioning

confidence: 99%

PRRGNVis: Multi-Level Visual Analysis of Comparison for Predicted Results of Recurrent Geometric Network

et al. 2022

View full text Add to dashboard Cite

The structure of a protein determines its function, and the advancement of machine learning has led to the rapid development of protein structure prediction. Protein structure comparison is crucial for inferring the evolutionary relationship of proteins, drug discovery, and protein design. In this paper, we propose a multi-level visual analysis method to improve the protein structure comparison between predicted and actual structures. Our method takes the predicted results of the Recurrent Geometric Network (RGN) as the main research object and is mainly designed following three levels of protein structure visualization on RGN. Firstly, at the prediction accuracy level of the RGN, we use the Global Distance Test—Total Score (GDT_TS) as the evaluation standard, then compare it with distance-based root mean square deviation (dRMSD) and Template Modeling Score (TM-Score) to analyze the prediction characteristics of the RGN. Secondly, the distance deviation, torsion angle, and other attributes are used to analyze the difference between the predicted structure and the actual structure at the structural similarity level. Next, at the structural stability level, the Ramachandran Plot and PictorialBar combine to be improved to detect the quality of the predicted structure and analyze whether the amino acid residues conform to the theoretical configuration. Finally, we interactively analyze the characteristics of the RGN with the above visualization effects and give reasons and reasonable suggestions. By case studies, we demonstrate that our method is effective and can also be used to analyze other predictive network results.

show abstract

Section: Protein Prediction-based Deep Learningmentioning

confidence: 99%