-
蛋白质模型质量评估方法是蛋白质结构预测的关键技术, 自CASP7以来一直是结构生物信息学领域的研究热点. 模型质量评估方法不仅可以指导蛋白质结构模型的精修, 还能够从多个候选构象中筛选出最佳模型, 具有重要的生物学研究和实际应用价值. 本文首先回顾了国际蛋白质结构预测关键评估竞赛(CASP)、全球持续蛋白质结构预测竞赛(CAMEO)以及单体蛋白和复合物的模型评估指标, 主要梳理了近5年来包括共识方法(多模型方法)、准单模型方法和单模型方法在内的模型质量评估方法的发展历程, 并介绍CASP15中的复合物模型评估方法; 鉴于深度学习在蛋白质预测领域所取得的巨大进展, 重点分析了深度学习在单模型方法数据集生成、蛋白质特征提取以及网络架构构建方面的深入应用, 并进一步介绍了本课题组近年来在模型质量评估方面开展的工作; 最后, 总结分析了目前蛋白质模型质量评估技术的局限性及所面临的挑战, 并对未来发展趋势进行了展望.The quality assessment of protein models is a key technology in protein structure prediction and has become a prominent research focus in the field of structural bioinformatics since advent of CASP7. Model quality assessment method not only guides the refinement of protein structure model but also plays a crucial role in selecting the best model from multiple candidate conformations, offering significant value in biological research and practical applications. This study begins with reviewing the critical assessment of protein structure prediction (CASP) and continuous automated model evaluation (CAMEO), and model evaluation metrics for monomeric and complex proteins. It primarily summarizes the development of model quality assessment methods in the last five years, including consensus methods (multi-model methods), single-model methods, and quasi-single-model methods, and also introduces the evaluation methods for protein complex models in CASP15. Given the remarkable progress of deep learning in protein prediction, the article focuses on the in-depth application of deep learning in single-model methods, including data set generation, protein feature extraction, and network architecture construction. Additionally, it presents the recent efforts of our research group in the field of model quality assessment. Finally, the article analyzes the limitations and challenges of current protein model quality assessment technology, and also looks forward to future development trends.
-
Keywords:
- protein model quality assessment/
- deep learning/
- single-model methods/
- complex model evaluation
[1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16] [17] [18] [19] [20] [21] [22] [23] [24] [25] [26] [27] [28] [29] [30] [31] [32] [33] [34] [35] [36] [37] [38] [39] [40] [41] [42] [43] [44] [45] [46] [47] [48] [49] [50] [51] [52] [53] [54] [55] [56] [57] [58] [59] [60] [61] [62] [63] [64] [65] [66] [67] [68] [69] [70] [71] [72] [73] [74] [75] [76] [77] [78] [79] [80] [81] [82] [83] [84] [85] [86] [87] [88] [89] [90] [91] [92] [93] [94] [95] [96] [97] [98] [99] [100] [101] -
Data sets URLs CASP https://predictioncenter.org/download_area/ CAMEO https://www.cameo3d.org/ Zhanglab https://zhanglab.ccmb.med.umich.edu/decoys/ AlphaFoldDB https://alphafold.ebi.ac.uk/ ESM Metagenomic Atlas https://esmatlas.com/resources?action=search_structure DeepAccNet https://github.com/hiranumn/DeepAccNet GNNRefine http://raptorx.uchicago.edu/download/ DeepUMQA https://academic.oup.com/bioinformatics/article/38/7/1895/6520805?login=true DeepUMQA3 https://www.biorxiv.org/content/10.1101/2023.04.24.538194v1.full.pdf+html GraphCPLMQA https://www.biorxiv.org/content/10.1101/2023.05.16.540981v1.full.pdf+html GraphGPSM https://academic.oup.com/bib/advance-article/doi/10.1093/bib/bbad219/7197734?searchresult=1#supplementary-data 
下载:
导出CSV
Predictor Name ROCnormalized PRnormalized Models $\rm AUC_{0,1}$ $ \rm AUC_{0,0.2}^*$ $\rm AUC_{0,1} $ $\rm AUC_{0.8,1}^*$ $\rm Received $ ZJUT-GraphCPLMQA 0.82 0.73 0.79 0.54 5143 DeepUMQA2 0.72 0.62 0.68 0.47 4468 DeepUMQA 0.73 0.60 0.67 0.45 4611 ModFOLD9 0.63 0.52 0.59 0.36 4309 QMEANDisCo3 0.9 0.66 0.79 0.49 6348 ProQ3D_LDDT 0.74 0.55 0.67 0.43 5171 QMEAN3 0.88 0.65 0.77 0.43 6348 ProQ3 0.72 0.53 0.66 0.39 5126 VoroMQA_v2 0.89 0.64 0.77 0.45 6350 ProQ2 0.86 0.59 0.74 0.39 6337 ProQ3D 0.70 0.47 0.61 0.35 5119 ModFOLD7_lDDT 0.84 0.53 0.69 0.41 6191 ModFOLD8 0.79 0.50 0.65 0.38 5802 Baseline Potential 0.80 0.51 0.66 0.32 6350 VoroMQA_sw5 0.82 0.50 0.65 0.36 6349 ModFOLD6 0.73 0.42 0.57 0.35 5380 下载:
导出CSV
Method Average TM-score Average Pearson Average bias GraphGPSM 0.730 0.633 0.126 MULTICOM_qa 0.485 0.715 0.258 ModFOLDdock 0.515 0.636 0.241 ModFOLDdockR 0.666 0.635 0.165 Venclovas 0.449 0.494 0.339 Manifold 0.582 0.541 0.179 Bhattacharya 0.387 0.474 0.361 *Real value 0.716 None None 注: *Real value 代表CASP15中所有蛋白质目标所有模型的真实平均TM-score分数. Note: *Real value represents the real average T-score of all targets in CASP15. 下载:
导出CSV
-
[1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16] [17] [18] [19] [20] [21] [22] [23] [24] [25] [26] [27] [28] [29] [30] [31] [32] [33] [34] [35] [36] [37] [38] [39] [40] [41] [42] [43] [44] [45] [46] [47] [48] [49] [50] [51] [52] [53] [54] [55] [56] [57] [58] [59] [60] [61] [62] [63] [64] [65] [66] [67] [68] [69] [70] [71] [72] [73] [74] [75] [76] [77] [78] [79] [80] [81] [82] [83] [84] [85] [86] [87] [88] [89] [90] [91] [92] [93] [94] [95] [96] [97] [98] [99] [100] [101]
下载:
计量
- 文章访问数:3419
- PDF下载量:199
- 被引次数:0