The decisive challenge in genetic diagnostic testing is the identification of genetic variants that are relevant for a medical diagnosis. For example, a patient’s exome, the subset of only 2% of the entire genome that encodes proteins, contains about 50,000 variants from which a clinician needs to identify the handful of potentially pathogenic variants that might be relevant for a diagnosis.
Limbus Medical Technologies GmbH operates the cloud-based software platform varvis® for processing and interpreting NGS data. The varvis® software enables, among other things, genetic laboratories to exchange the results of their analyses in an automated fashion. This provides genetic laboratories with urgently needed reference data to interpret the clinical relevance of their results, and thereby significantly improves the diagnostic yield of genetic analyses and the quality of the diagnoses based on them.
The development of varvis® also presents some novel scientific challenges in terms of modeling genetic knowledge that is distributed and rapidly changing in laboratories. Another challenge is the automated processing and analysis of genetic data. Ideally, any newly discovered and relevant data should be considered immediately for the variant interpretation of the following cases.
An AI system for variant interpretation that is integrated into the reference data network should therefore adapt to incoming data automatically, and thus needs to draw on concepts from meta-learning and incremental machine learning.
The goal of the IDEA-PRIO project
The process of variant interpretation is time-consuming and requires expert knowledge. Therefore, the most urgent task so far is to extend the data network with a variant prioritization mechanism that automatically exploits the available (but distributed) data and thus enables the physician to prioritize variants based on a large amount of data collected in other laboratories. This will dramatically improve diagnostic yield, greatly reduce the time required for the diagnostic process, and shorten test turnaround time.
To ensure the safety of providing an AI-based variant interpretation, the project is also focused on the application of explainable AI methods. This means that a physician will be able to follow and understand the reasoning behind an AI-based variant interpretation. Additional challenges include documenting the input data provenance and safeguarding the operational integrity of such a complex service by employing techniques like runtime verification.
Der Forschungsverbund gratuliert zu diesem Erfolg und wünscht auch weiterhin gutes Gelingen!