gms | German Medical Science

Introduction: Personalized medicine is a rapidly evolving field that considers individual variations of patients in disease prevention, health monitoring, and treatment. This trend was accelerated by a growing use and maintenance of electronic health records (EHRs) and the increasing availability of biomedical data such as gene panels paving the way for more precise and effective treatments for each patient.

However, handling large amounts of multidimensional clinical and genomic data can be a challenging task.

State of the art: To keep up with the ever growing volume of data the need for the development of interactive tools for data-exploration is tremendous.

Applications like cBioPortal [1] allow their users to analyze, browse and visualize high dimensional clinical data. Moreover, the exploration of clinically similar patient-groups based on multi-dimensional data has shown to add crucial value for researchers and potentially clinicians [2], [3].

However, to our best knowledge the community is lacking a software-package that combines clinical and genomic patient-data that aid the discovery of such similar patient-groups.?????

Concept: To address this, we developed a prototype that enables researchers and clinicians to explore and analyze large amounts of multi-dimensional patient-data in an easily understandable and interpretable way. The software package integrates local data as well as real-world EHRs and gene expression data that can be imported from cBioPortal. In particular, we utilize patient-features, such as for instance, demographic data like age and sex, diagnoses similarity, and combine these with expert derived panel sequencing information. The package enables users to interactively choose weights for the patient-features to allow clinicians and researchers to adjust the analysis to their needs.

Implementation: We implemented the prototype using python's Dash library. We integrated gene enrichment analysis using the PANTHER API [4]. In addition to our in-house data we use publicly available data that we derived using the cBioPortal API. Genetic- as well as diagnostic-similarity are being calculated using semantic-similarity for the gene ontology and the ICD-10 taxonomy. Those features are combined in a patient-vector with other demographic and clinical attributes like age, at at diagnoses and sex.

Our software then visualizes patient-similarities by plotting e.g. 2D-distances and heatmaps of the calculated similarity scores and allows users to interactively choose the weights of each available patient-feature. ????Furthermore, the users have the option to filter the genes that are included into the calculation regarding their clinical relevance in relation to cancer using the OncoKB API [5].

Conclusion: By clustering patients according to their clinical phenotypic and transcriptomic profiles, the package allows for interactive analyses and identification of similar patient-groups. These similarity groups can help researchers and healthcare providers to identify patients who are more likely to show similar outcomes, or potential side effects of certain treatments.

Ultimately the developed prototype has the potential to pave the way for clinical decision support systems that can improve patient-outcomes and enable healthcare providers making informed treatment decisions based on a more comprehensive understanding of patient-characteristics.

In the future more patient-features could be included to further extend the functionality of the tool.??????

The authors declare that they have no competing interests.

The authors declare that a positive ethics committee vote has been obtained.