Datensatz

Zusammenfassung

Hepatitis B is one of the most frequent infectious diseases worldwide. Thus improvement of therapy is an active area of research. Since the hepatitis B virus can adapt to treatment in terms of resistance mutations, therapeutic approaches often fail. With computational support and statistical learning methods, it is possible to adjust therapy according to the underlying virus. In consequence, an improvement of therapy is enabled due to individual treatment. In this thesis, an approach to predict the resistance from genome sequence is presented. Resistance is defined in terms of viral IC-50 concentration. A special feature described is the usage of sample specific weights based on the precision of the phenotypic assay. These weights are meant to overcome a distortion of prediction results caused by experimental measurement errors. For this purpose, the LASSO and linear support vector machines were trained with 367 hepatitis B clones treated with four approved antiviral drugs. Using uncertainty estimates, it was possible to moderately improve prediction of the IC50 value from genotype for two out of four drugs. In summary, this thesis provides a recent approach to support individual treatment by improving prediction of the IC-50 value from viral genotype using sample specific weights.