Protein Structure Classification and Loop Modeling Using Multiple Ramachandran Distributions

Seyed Morteza Najibi, Mehdi Maadooliat, Lan Zhou, Jianhua Z. Huang, Xin GaoComputational 2017 · DOI

TL;DR

This paper introduces a novel statistical method, PSCDE(T), for analyzing protein structures by modeling their backbone angles. It offers a faster and more efficient way to classify protein structures and improve loop modeling compared to previous approaches, by leveraging trigonometric splines to better handle the circular nature of the data.

What Problem Does This Paper Address?

Proteins fold into complex three-dimensional structures that are crucial for their function. Understanding these structures is a major goal in biology. A key aspect of protein structure is the conformation of its backbone, which can be described by dihedral angles (phi and psi). These angles have a circular nature, meaning that 0 degrees is the same as 360 degrees. Traditional statistical methods struggle with this circularity.

The Ramachandran plot is a common visualization tool that plots phi vs. psi angles, revealing allowed conformational regions. However, efficiently modeling and comparing these plots across many proteins, especially when dealing with limited data or subtle differences, is challenging. Existing methods for analyzing these angular distributions often require separate modeling for each protein or group, which can be computationally expensive and less accurate when data is sparse. This paper aims to address these limitations by developing a more sophisticated and efficient statistical framework for collective density estimation of these angular distributions, which can then be applied to protein classification and loop modeling.

Methods

The core of this paper is the development of a new statistical method called PSCDE(T) (Penalized Spline Collective Density Estimation with Trigonometric splines). This method builds upon a previous approach (PSCDE) but introduces several key improvements:

  • Trigonometric Splines: Instead of using standard B-spline basis functions that require artificial constraints to handle circular data, PSCDE(T) employs trigonometric B-spline basis functions. These functions are inherently periodic and smooth, naturally accommodating the circular nature of protein backbone angles without complex workarounds.
  • Tensor Product Approach: The method uses a tensor product of these trigonometric splines to create bivariate basis functions, which are well-suited for modeling the two-dimensional Ramachandran plots.
  • Efficient Smoothing Parameter Selection: A significant bottleneck in previous methods was the time-consuming grid search required to find the optimal smoothing parameter (lambda). PSCDE(T) adopts an iterative approach, updating the smoothing parameter within the Newton-Raphson optimization algorithm. This dramatically speeds up the process without sacrificing accuracy.
  • Collective Density Estimation: The method estimates density functions for multiple groups of protein angles simultaneously. By assuming a common set of basis functions across these groups, it "borrows strength" from similar distributions, leading to more robust and accurate density estimates, especially in data-sparse regions.
  • Low-Dimensional Representation: The coefficients of the fitted basis functions provide a compact, low-dimensional representation of each protein's angular distribution. These coefficients can then be used as features for downstream tasks like clustering and classification.

Datasets

The paper uses datasets derived from the Structural Classification of Proteins (SCOP) database. SCOP is a widely used resource that manually classifies known protein structures based on their evolutionary, structural, and functional relationships.

The authors designed four specific tasks to evaluate their method, categorized by difficulty:

  • SCOP.1 (Easy Task): 63 protein domains from three remote Protein Classes.
  • SCOP.2 (Somewhat Hard Task): 33 domains from four species within the same protein subclass (all-alpha protein class). This task is harder because the domains are very similar, differing only at the lowest levels of the SCOP hierarchy.
  • SCOP.3 (Hard Task): 40 protein chains from three different Fold/Superfamily levels within the "Alpha and beta proteins" class.
  • SCOP.4 (Challenging Task): 26 protein chains from the "All beta proteins" class, taken from a challenging benchmark dataset designed to assess fold recognition methods.

Protein Data Bank (PDB) files were used to obtain atomic coordinates, from which the (phi, psi) angles were derived using the R package PRESS.

Key Insights and Ideas

  • Trigonometric splines are a natural fit for circular data: The paper elegantly demonstrates how using inherently periodic trigonometric splines simplifies the modeling of protein backbone angles, avoiding the need for complex boundary handling required by other methods.
  • Collective estimation is powerful for sparse data: By sharing information across multiple protein angle distributions, the method becomes more robust when dealing with limited data, a common issue in biological datasets.
  • Efficient parameter tuning is crucial for practical application: The speedup achieved by updating the smoothing parameter within the optimization loop is a significant practical advantage, making the method more accessible for large-scale analyses.
  • Low-dimensional coefficients offer a new perspective for classification: The coefficients derived from the fitted densities provide a concise feature set that can be used for alignment-free protein classification, offering a unique alternative to traditional sequence or structure alignment methods.

Key Results

  • Improved Protein Classification: PSCDE(T) demonstrated competitive or superior performance in protein structure classification tasks across various difficulty levels compared to established methods like Needleman-Wunsch, Smith-Waterman, TM-align, Dali, KDE, and the original PSCDE. Notably, it performed very well on the "Challenging Task" (SCOP.4), achieving high NMI and ARI scores.
  • Significant Speedup: The PSCDE(T) method was substantially faster than the original PSCDE, with running times reduced by up to an order of magnitude (e.g., 1.39 min vs. 14.85 min for SCOP.2). This is attributed to the efficient updating of the smoothing parameter.
  • Accurate Density Estimation: The method successfully estimated neighbor-dependent Ramachandran distributions, which were then used to improve loop modeling accuracy in conjunction with the Rosetta software.

Limitations and Future Work

  • Summary Statistics: The authors acknowledge that angular density is a summary statistic and may ignore other important aspects of protein structure. Therefore, it might not always yield the best results for every arbitrary dataset.
  • Complementary Approach: They suggest that PSCDE(T) can be used independently or as a complement to existing methods.
  • Future Work: The paper mentions the development of a web application and a toolbox to facilitate the use of PSCDE(T) for collective estimation of Ramachandran distributions in various applications, including loop modeling and rotamer library generation. The method's applicability to higher-dimensional density estimation is also highlighted as a potential future direction.

Summary updated

Abstract

Recently, the study of protein structures using angular representations has attracted much attention among structural biologists. The main challenge is how to efficiently model the continuous conformational space of the protein structures based on the differences and similarities between different Ramachandran plots. Despite the presence of statistical methods for modeling angular data of proteins, there is still a substantial need for more sophisticated and faster statistical tools to model the large-scale circular datasets. To address this need, we have developed a nonparametric method for collective estimation of multiple bivariate density functions for a collection of populations of protein backbone angles. The proposed method takes into account the circular nature of the angular data using trigonometric spline which is more efficient compared to existing methods. This collective density estimation approach is widely applicable when there is a need to estimate multiple density functions from different populations with common features. Moreover, the coefficients of adaptive basis expansion for the fitted densities provide a low-dimensional representation that is useful for visualization, clustering, and classification of the densities. The proposed method provides a novel and unique perspective to two important and challenging problems in protein structure research: structure-based protein classification and angular-sampling-based protein loop structure prediction.
BibTeX
@inproceedings{najibi2017protein,
  author = {Seyed Morteza Najibi and Mehdi Maadooliat and Lan Zhou and Jianhua Z. Huang and Xin Gao},
  title = {{Protein Structure Classification and Loop Modeling Using Multiple Ramachandran Distributions}},
  booktitle = {Computational},
  year = {2017},
  doi = {10.1016/j.csbj.2017.01.011},
}
Figure