Unlocking Predictive Power: A Machine Learning Tool Derived from In-Depth Analysis to Forecast the Impact of Missense Variants in Human Filamin C

Abstract Cardiomyopathies, diseases of the heart muscle, are a leading cause of heart failure. An increasing proportion of cardiomyopathies have been associated with specific genetic changes, such as mutations in FLNC , the gene that codes for filamin C. Altogether, more than 300 variants of FLNC have been identified in patients, including a number of single point mutations. However, the role of a significant number of these mutations remains unknown. Here, we conducted a comprehensive analysis, starting from clinical data that led to identification of new pathogenic and non-pathogenic FLNC variants. We selected some of these variants for further characterization that included studies of in vivo effects on the morphology of neonatal cardiomyocytes to establish links to phenotype, and the in vitro thermal stability and structure determination to understand biophysical factors impacting function. We used these findings to compile vast datasets of pathogenic and non-pathogenic variant structures and developed a machine-learning-based neural network (AMIVA-F) to predict the impact of single point mutations. AMIVA-F outperformed most commonly used predictors both in disease related as well as neutral variants, approaching ∼80% accuracy. Taken together, our study documents additional FLNC variants, their biophysical and structural properties, and their link to the disease phenotype. Furthermore, we developed a state-of-the-art web-based server AMIVA-F that can be used for accurate predictions regarding the effect of single point mutations in human filamin C, with broad implications for basic and clinical research.