• Statistical analysis in Spectrolyzer
• Available methods and algorithms
• Examples
• Videos

• You can import your data from CSV files, you can also use MALDI or LC-MS data directly for preprocessing.
• We give you two ways of sorting through your data: classical and state-of-the-art. Whether you are interesting in discovering more or using proven methods, Spectrolyzer is for you.
• Potential biomarkers (i.e. feature discriminating between two groups, e.g. Healthy vs. Disease, Treated vs. Control, or any other two-group-comparison) can be detected with higher precision and accuracy than in your existing software.
• Various classification methods can be applied combined with suitable approaches to feature selection
• The discovery of interesting peptides or molecules (discriminative features) can be followed by database searches for protein and metabolites identification with the help of either Mascot, X!Tandem or METLIN batch search.

Statistical analysis module contains advanced statistical and data analysis methods (e.g. PCA, classification models, statistical tests) and provides very powerful support for biomarker discovery.

One-dimensional data exploration

• Examining basic properties of a single (selected) feature (including interactive visualization).
• Popular summary statistics and classical statistical tests.
• Visualization with standard graphical tools.

Multi-dimensional data exploration

• Finding relationships (similarities) between samples or between features.
• Data visualization using heat maps.

Dimension reduction

• Feature extraction algorithms, e.g. PCA (Principal Component Analysis),
• Graphical (2D and 3D) and analytical tools for quality of data reduction assessment.
• Advanced analysis techniques, e.g. Partial Least Squares combined with Logistic Regression (PLS-LR).

Single biomarker detection

• Examining all the features for discriminative properties.
• Creating and displaying ranking of features.
• Popular separation measures available, including, e.g. t-test and AUC. 

Multiple biomarker detection 

• Finding a robust subset of the most discriminative features (candidate biomarkers) that can be expected to give reproducible results for new experiments.
• Advanced feature ensemble approach to find stable features yielding accurate classification.
• The performance of classification model assessed in detail in order to find a good balance between classifier accuracy and complexity.
• Tools facilitating selection of an optimal number of best features.

Classification for building predictive models

• Popular and efficient classification algorithms including, e.g. LDA, PLS, Logistic Regression, Decision Tree, randomForest, SVM (Support Vector Machines), k-NN.
• Comprehensive and flexible framework for classifier performance evaluation.
• Optional stages in classifier performance evaluation.
• Opportunity to choose from different evaluation schemes e.g. simple learning/test split, k-fold cross-validation, leave-one-out, random subsampling.
• Feature selection algorithms, e.g. filter-based feature selection and stepwise feature selection methods (Forward/Backward Feature Selection).
• A number of classification performance measures, including misclassification error estimation and ROC analysis.
• Ranking of selected features (i.e. features resulting in best classification results).

Cluster analysis

• Popular clustering algorithms, e.g. k-means, agglomerative clustering.
• Cluster validation tools.
• Automatic choice of optimal number of clusters.
• Graphical tools facilitating interpretation of clustering results.