Variants identify sarcomere inter-protein contacts distinguishing inheritable cardiac muscle diseases

Human ventriculum myosin ({beta}mys) powers contraction sometimes while complexed with myosin binding protein C (MYBPC3) on the myosin thick filament. The latter regulates {beta}mys activity through inter-protein contacts. Single nucleotide variants (SNVs) change protein sequence in {beta}mys or MYBPC3. They cause inheritable heart disease. When a SNV modified domain locates to an inter-protein contact it affects complex coordination. Domains involved, one in {beta}mys and the other in MYBPC3, form coordinated domains called co-domains. Co-domains are bilateral implying the potential for a shared impact from SNV modification in either domain suggesting their joint response to a common perturbation assigns location. Human population genetic divergence is the common systemic perturbation. A general contraction model with a neural/Bayes network design reveals SNV probabilities specifying correlations between domain members using 2D correlation genetics (2D-CG). It reveals co-domain locations in three common human heart diseases caused by SNVs, familial hypertrophic cardiomyopathy (FHC), dilated cardiomyopathy (DCM), and left ventricle non-compaction (LVN). Co-domain maps for DCM and LVN link MYBPC3 with two levels of myosin heads on the myosin thick filament surface implying these myosin dimers form the super-relaxed state (SRX). The FHC co-domain map involves just one myosin dimer implying the myosins do not form SRX. Comparing co-domain maps for FHC, DCM, and LVN phenotypes suggests SRX disruption involves a co-domain between MYBPC3 regulatory domain and the myosin regulatory light chain (RLC) N-terminus. The general contraction model scenarios, constructed from feed-forward neural networks, were explored with the purpose to understand how to interpret them mechanistically with basic natural language characteristics. These characteristics emerge from dependencies among inputs coded in hidden layer width and depth when they are deciphered using 2D-CG. In this application, the thick filament structural states emerge for FHC, DCM, and LVN phenotypes defining thick filament structural state joining the other standard characteristics of phenotype and pathogenicity. Emergent natural language interpretations for general network contraction models are on the horizon.