< Back to Article

Genome Modeling System: A Knowledge Management Platform for Genomics

Fig 2

Key concepts of the GMS.

The genome modeling system is architected around the idea of a ‘genome model’. The following vignettes illustrate key concepts integral to these models: (A) A subject can be modeled multiple times, possibly each with distinct ‘processing profiles’. For example, two different models can be defined for the HCC1395 genome using the ‘reference alignment’ pipeline. In Model 1, the processing profile specifies the use of BWA for alignment and Samtools for variant detection. In Model 2, Bowtie2 and GATK are used for these steps instead. (B) A given processing profile can be used across a group of models, ensuring, for instance, that all subjects in a cohort are processed in similar ways. In this example, two different cell line genomes (HCC1395 and XY2123) have models defined of the exact same type, using the processing profile with BWA/Samtools specified. (C) A model has no results until a build is generated. If the model is updated to have new inputs, a new build is required. Builds are immutable snapshots of modeling pipeline results. In this example, the HCC1395 genome has a reference alignment model again making use of the BWA/Samtools profile. However, as new instrument data becomes available, new builds are constructed to reflect the most complete data. (D) When models are used as inputs for other models, the last complete build for the input model is used as an input for the downstream build. In this example, both tumor and normal genomes are available for an individual (in this case HCC1395). Reference alignment models are built for each sample and then both are used as inputs for a third ‘somatic variation’ model. In reality, it is the underlying data in the reference alignment builds that are used to create a somatic variation build, identifying all variants that are thought to be tumor specific.

Fig 2