Generating quantitative models describing the sequence specificity of biological processes with the stabilized matrix method.

Peters B, Sette A. · 2005

Read Generating quantitative models describing the sequence specificity of biological processes with the stabilized matrix method. by Peters B, Sette A. (2005) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/europmc-pmc1173087.txt.

Introduction

Peters and Sette present a software implementation of their stabilized matrix method in this 2005 paper. Many molecular processes recognise short sequences, such as immune peptides binding to MHC molecules, the display platforms that show cell contents to the immune system. From experimental data, a model of that sequence specificity can be built as a motif, a scoring matrix, or a neural network.

Such models summarise results and predict outcomes for untested sequences. Their method, the stabilized matrix method, had already predicted peptide binding to MHC, peptide transport by the TAP transporter, and proteasomal cleavage of proteins. The proteasome is the cell's protein shredder. Here the algorithm is released as publicly available software so other labs can build models the same way. The authors describe the implementation and its intended uses.

Catalog record

CREATORPeters B, Sette A.
YEAR2005
COLLECTIONPAPERS
FORMATTXT
FILE SIZE34 KB
LICENSECC BY 4.0
RIGHTSCC BY
SOURCEEurope PMC

Source: https://europepmc.org/articles/PMC1173087

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Generating quantitative models describing the sequence specificity of biological processes with the stabilized matrix method By Bjoern Peters, Alessandro Sette Published in: BMC Bioinformatics DOI: 10.1186/1471-2105-6-132 ABSTRACT Abstract Background Many processes in molecular biology involve the recognition of short sequences of nucleic-or amino acids, such as the binding of immunogenic peptides to major histocompatibility complex (MHC) molecules. From experimental data, a model of the sequence specificity of these processes can be constructed, such as a sequence motif, a scoring matrix or an artificial neural network. The purpose of these models is two-fold. First, they can provide a summary of experimental results, allowing for a deeper understanding of the mechanisms involved in sequence recognition. Second, such models can be used to predict the experimental outcome for yet untested sequences. In the past we reported the development of a method to generate such models called the Stabilized Matrix Method (SMM). This method has been successfully applied to predicting peptide binding to MHC molecules, peptide transport by the transporter associated with antigen presentation (TAP) and proteasomal cleavage of protein sequences. Results Herein we report the implementation of the SMM algorithm as a publicly available software package. Specific features determining the type of problems the method is most appropriate for are discussed. Advantageous features of the package are: (1) the output generated is easy to interpret, (2) input and output are both quantitative, (3) specific computational strategies to handle experimental noise are built in, (4) the algorithm is designed to effectively handle bounded experimental data, (5) experimental data from randomized peptide libraries and conventional peptides can easily be combined, and (6) it is possible to incorporate pair interactions between positions of a sequence. Conclusion Making the SMM method publicly available enables bioinformaticians and experimental biologists to easily access it, to compare its performance to other prediction methods, and to extend it to other applications. BACKGROUND Whenever experimental data is gathered to examine a process, researchers will try – implicitly or explicitly – to define a model describing the process. The purpose of building a model is to generalize the experimental data, allowing the prediction of new experimental outcomes, or to gain insights into the experimental process. In the biological sciences, models are often implicitly built and verbally formulated, such as "the proteasome preferably cleaves after hydrophobic amino acids". Such a model is easy to understand and often reflects all the knowledge that can be gathered from a few, difficult experiments. However, with the advent of high-throughput experiments to the biosciences, it has become feasible to generate more quantitative, mathematical models, based on large volumes of data. For the purpose of building a model, conducting experiments can be formally described as collecting pairs of experimental parameters x and experimental outcomes (measurements: ymeas). Building a model is then equivalent to finding a function f(x) = ypred ≈ ymeas. Herein, we call the model function f a prediction tool, and the experimental observations used to generate it its training set T = (x, ymeas). Developing a prediction tool is a two step process. First, a general prediction method capable of generating specific tools has to be chosen, such as a certain type of neural network, classification tree, hidden markov model or regression function. This choice is to some degree determined by the experimental data available. However, only a few prediction methods are clearly unsuited for a certain type or amount of data, leaving several potentially appropriate methods to choose from. In practice, the personal experience of a scientist often determines which prediction method he applies, as learning to apply a new method is often more costly than the benefits of a slightly better model. After a method is chosen, it is applied to the experimental data to generate a specific prediction tool. Several different terms are commonly used for this second step, such as 'supervised learning', 'fitting the model' or 'regression'. Each method has its own formalism describing how this is done, but essentially a method is capable of generating a certain class of functions, of which one is chosen, that minimizes the difference between measured and predicted experimental outcomes for the training set T. Here we describe a computer program implementing the SMM prediction method, which can be applied to model the sequence specificity of quantifiable biological processes. In such experiments, the input parameter × corresponds to a sequences of amino-or nucleic acids, and the experimental outcome ymeas is a real numbers measuring the process efficiency. The method utilizes training sets in which the examined sequences all have the same length. The SMM method was previously applied successfully to predictions of MHC binding [1], TAP transport [2] and proteasomal cleavage [3]. However, its software implementation was not made publicly available, because it relied on a commercial library with restrictive licensing terms. Also, applying the method to a new problem required manual changes in the source code, which is impracticable for an external user. Both of these issues are addressed in the software implementation made available here. In addition, this manuscript describes two specific features that we have found to be effective in generating high quality models and which can easily be utilized in other prediction methods, namely handling of bounded data and combinatorial peptide libraries. The SMM software is not meant for 'classical' sequence analysis, which can be roughly defined as aligning related sequences in order to identify conserved residues or in order to generate classifiers which can identify additional related sequences. Rather, the typical application is the characterization of a sequence recognition event, such as the sequence specific cleavage efficiency of a protease. This characterization does not assume any evolutionary relationship between different recognized sequences. In terms of scientific fields, that makes the SMM software aimed at applications in biochemistry or molecular biology. IMPLEMENTATION Code The SMM algorithm is implemented in C++ code. Only standard libraries or freely available external libraries were used. The two external libraries are Tinyxml [4] to handle the XML input and output and the Gnu Scientific Library [5] for efficient vector and matrix operations. The source code has been compiled and tested using Visual C++ on a windows system and using g++ (1.5 or above) on a Debian and Suse Linux system. A windows executable is also available. The source code, documentation and examples are available as additional files 1 and 2 of this manuscript, and on the project homepage at http://www.mhc-pathway.net/smm. On a standard 2.6 GHz Pentium 4 PC running windows XP, the creation of a prediction tool from a typical set of training data containing 300 8-mer peptides takes about 10 minutes. Input and output The program expects an XML document as its standard input. A more sophisticated (graphical) user interface is not likely to be of great interest for the projected user community, most of which will probably call the executable file from a script or integrate the code into an existing program. The XML input document contains either training data to generate a new tool (Figure 1) or sequences for which a prediction should be made with a previously generated tool. The output of the program is again in XML format. For all types of input and output, XML schemas defining their exact structure are supplied.

Read the complete text