Diagnostic algorithms - VeViVas customized expert teams

Go to content

Diagnostic algorithms

Data analysis
placeholder

A proprietary method to develop robust
diagnostic algorithms



Performant biomarkers are essential for modern diagnostics and personalized medicine.
However, the identification of markers that provide reliable results is challanging for a number of reasons:

Heterogeneous populations

While a group of patients may share clinical features that result in a common diagnostic label, the underlying pathogenic mechanisms may be rather diverse. As a consequence, at the molecular level, these patients may fall into different subgroups that cannot be identified by the same biomarkes.

The overfitting dilemma

The wealth of molecular information provided by "-omics" technologies poses a statistical challenge, as the number of samples that are available for the discovery of biomarkers is typically far smaller than the number of analytes. This frequently leads to a situation where candidate markers from the discovery sample set cannot be confirmed in an independent patient cohort.
Variability of sample handling

Even if sample processing protocols and analytical devices are optimized for reproducibility, some "batch effects" typically remain for samples that are processed at different locations, at different times and by different people. This leads to experimental noise and hence a loss of precious diagnostic information.
The VeViVas Dx procedure to identify robust diagnostic markers from "-omics" datasets is based on more than ten years of experience analyzing transcriptome data of various patient populations from pre-clinical and clinical studies. The method employs proprietary mathematical procedures which allow to identify markers that :

I)     behave consistently in independent patient cohorts

II)    are robust against technical variability

III)   are suitable to detect previously unknown patient subpopulations

Examples
Psoriasis versus healthy skin
Affymetrix transcriptome data of 27 different public datasets (GEO) were used to generate an algorithm that distinguishes between healthy skin and lesional psoriasis. The algorithm reliably classified the samples of the test set (not used for training the algorithm).
Psoriasis prediction scores for skin samples (circles) in training and test set
Diagnostic test performance

Psoriasis versus atopic dermatitis
Affymetrix transcriptome data of 26 different public datasets (GEO) were used to generate a performant algorithm that distinguishes between lesional psoriasis and lesional atopic dermatitis (skin biopsies).
Atopic dermatitis prediction scores for skin samples (circles) in training and test set
Diagnostic test performance

Blood samples of breast cancer
Analysis of the GSE27562 dataset (GEO) of PBMC samples demonstrates that the method is also applicable for sample types with highly variable cellular composition. Even though this is a small dataset, training and test set show very consistent behavior.
Diagnostic test performance
Breast cancer prediction scores for PBMC blood samples (circles) in training and test set

Whole blood samples of
rheumatoid arthritis
Analysis of the GSE93777 dataset (GEO) shows very good performance of the test in a highly variable tissue (whole blood) in a disease which is known to consist of poorly defined sub-populations.
Rheumatoid arthritis prediction scores for whole blood samples (circles) in training and test set
Diagnostic test performance
    CO2 - neutral     VeViVas uses CO2-emission-free hosting services, electricity and heating      
Subject to technical changes and printing errors. Copyright (c) 2020 VeViVas

Back to content