SoMaCX: a complex generative genome modeling framework.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Somatic structural variations (SVs) are prevalent in cancer but difficult to detect due to tumor heterogeneity and sequencing limitations.
• Detecting SVs requires a sufficient read fraction, and complex events like chromothripsis pose interpretation challenges.
• In vivo measurement of SVs is difficult, necessitating realistic simulation frameworks to understand system limitations.

Purpose of the Study:

• To develop a generative modeling approach for simulating somatic structural variations.
• To create a framework that addresses limitations in current SV detection methods.
• To generate realistic simulated genomic data for evaluating SV calling algorithms.

Main Methods:

• Developed soMaCX, a generative framework utilizing data distributions for realistic simulations.
• Incorporated mechanisms for germline conservation, somatic tissue composition, and regional distribution controls.
• Enabled complex SV generation, outputting FASTA files compatible with various downstream read simulators (e.g., Illumina, PacBio).

Main Results:

• soMaCX demonstrates superior generative modeling performance compared to existing simulation frameworks when assessed against real data.
• The framework generates FASTQ and BAM files for SV calling, supporting multiple sequencing technologies.
• Simulated data can be used to assess germline variation calling performance and calibrate for rare variants.

Conclusions:

• The soMaCX framework offers a more realistic and detailed simulation of genomes by using biologically relevant regions and pathways.
• This open-source tool enhances the evaluation of SV detection methods, particularly for challenging genomic conditions.
• soMaCX aids in measuring germline variation calling performance and calibrating for clinically relevant rare variants.