Hepatic steatosis (fatty liver) is a severe liver disease induced by the excessive accumulation of fatty acids in hepatocytes. Factors which may provoke the onset of steatosis are manifold: excessive alcohol consumption, obesity, diabetes, or mitochondrial dysfunction. Aggravation of the disease (persistence of provoking factors) typically leads to steatohepatitis, liver cirrhosis and liver cancer. In this study we aim to build reliable in-silico models for predicting hepatic steatosis on basis of in vivo data. After careful data curation, the data set of 1121 compounds (131 steatosis positives, 990 steatosis negatives) was split into a training set (70 %) and an external test set (30 %) by keeping the initial distribution of actives/inactives of the original data set. We used different machine learning techniques such as Random Forest, Support vector machine, Artificial Neural Networks and NaiveBayes in combination with different sets of molecular descriptors (physiochemical properties, Toxprint fingerprints, RDKit fingerprints) in order to identify models with high predictive power. In addition, the influence of the incorporation of different sets of biological fingerprints – such as predictions from in vitro steatosis models - was systematically investigated. Further, we used techniques such as cost sensitive learning, stratified bagging and conformal prediction in order to handle the data imbalance. In summary, we achieved reliable computational models for predicting hepatic steatosis based on in vivo data.

Machine learning models for predicting hepatic steatosis based on in vivo data

256th National Meeting (2018)

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.