FAST-AIMS

Category Proteomics>Mass Spectrometry Analysis/Tools

Abstract FAST-AIMS (Fully Automated Software Tool for Artificial Intelligence in Mass Spectrometry) is a Clinical Mass Spectrometry Analysis System.

The system is a stand-alone application that runs on a Windows computer and can work with mass spectrometry (MS) data generated by different instruments, for example, matrix-assisted laser desorption/ionization (MALDI), surface-enhanced laser desorption/ionization (SELDI), liquid chromatography tandem mass spectrometry LC-MS/MS, etc.

The FAST-AIMS system consists of a command-line executable (implemented in MATLAB and C) and a wizard-like graphic user interface (implemented in Borland Delphi) that calls the above executable with different parameters.

It requires neither MATLAB nor other software to be installed. The need for MATLAB is circumvented by using the MATLAB Compiler that translates all relevant MATLAB code into C.

FAST-AIMS possesses an intuitive ‘wizard-like’ interface with defaults provided so that users need Not be familiar with all steps of data analysis, while more experienced analysts may benefit from the command-line interface for batch analyses.

Given a mass spectrometry dataset as input, FAST-AIMS can automatically perform one of the following tasks:

1) Generate a classification model by optimizing the parameters of the analysis algorithms to ensure its ‘optimal performance’;

2) Obtain an unbiased estimate of the future ‘classification performance’ of the ‘optimized model’;

3) Generate a model and estimate classification performance in tandem; and

4) Apply an ‘existing model’ to a new set of samples.

During the process, the system also offers the option of identifying ‘biomarkers/features’ that capture the classification tasks of interest and can be used to explore underlying biological mechanisms.

All FAST-AIMS functions follow the nested ‘N-fold cross-validation’ design that allows optimization of ‘predictive model’ parameters while providing an unbiased classification performance estimate.

The following is an outline of the main steps in the analysis process as performed by FAST-AIMS:

1) An opening (Windows) window asks whether the user wishes to start a new analysis or load a previously saved analysis. In starting a new analysis, the user first selects the location of a mass spec data file.

The system recognizes space delimited text data files, each row of which corresponds to the ‘intensity values’ of a sample and the first column of which corresponds to the response/outcome of each sample.

The user can optionally specify the location of the text file with mass/charge (M/Z) values arranged in a column. The latter file is used for preprocessing of mass spectrometry data and for the generation of the analysis report.

2) The user is then asked to choose from one of the four (4) analysis tasks outlined above.

3) Next, the user selects ‘cross-validation design’: N-fold cross-validation or Leave-one-out cross-validation.

4) Then, the user is asked to select whether ‘baseline subtraction’ and ‘peak detection’ and/or peak alignment should be performed.

5) The user specifies a ‘normalization sequence’ to be applied to the data.

6) The classifier and its parameters are selected by the user. The system uses Support-Vector Machine (SVM) classification algorithms because of their high performance in published analyses of mass spectrometry data and other types of high-throughput data, most notably ‘gene expression’ arrays in which SVMs outperformed all major ‘pattern recognition’ algorithms.

7) Feature/variable selection algorithms (and their parameters) are then selected by the user. The algorithms include: using all features in the data, HITON-PC, and SVM weight-based feature selection.

8) The user then selects a ‘performance estimation’ metric: either area under Receiver Operating Characteristics (ROC) curve, accuracy, or ‘entropy-based’ metric relative classifier information (RCI).

9) A ‘log file’ destination is specified by the user.

10) A ‘report file destination’ is specified by the user.

11) The user then clicks “Run”.

If the task involves the generation of a ‘classification model’ and/or performance estimation, the system considers each permutation of the classifier and ‘feature selection’ algorithm parameters.

Some steps can be performed on each spectrum independently (e.g., peak detection, baseline subtraction, and some normalization methods).

All steps that require consideration of multiple spectra (e.g., peak alignment, classification, and feature selection) are performed on each training set of the data to ensure unbiased estimates of the classification performance.

The FAST-AIMS system is designed for mass spectrometry data analysis and inherits challenges of using that technology.

For example, given a mass spectrometry dataset a system identifies biomarkers/features that correspond to anonymous peak intensities and may have little significance for the clinical practice.

Any potentially interesting peak would have to be examined using an MS/MS system and the findings have to be confirmed by a quantitative proteomics approach.

However, the manufacturers emphasize that FAST-AIMS can be universally applied to various MS and tandem MS ‘measurement technologies’ and does Not require use of specific proteomic platforms such as SELDI or MALDI.

System Requirements

Web-based.

Manufacturer

Manufacturer Web Site FAST-AIMS

Price Contact manufacturer.

G6G Abstract Number 20557

G6G Manufacturer Number 104166