Fermentation Seminar Department of Biochemistry and Microbio logy

Studies trying to uncover molecular disease patterns show that in many cases specific amino acid substitutions can be linked to known disease pathways. Thus it is not surprising that we have quite a large repertoire of methods at hand to asses the impact of mis-sense mutations, often caused by single nucleotide polymorphisms (SNPs). However, comparing method performance, most of those methods perform very similar. Despite of an increase in number and complexity of prediction methods, prediction accuracy seems to not have changed significantly in the last two decades. In contrary, prediction accuracy has slowly reached a threshold which is by far not sufficient for wide range application, i.e. in diagnostics. We have analyzed key aspects for this behavior and recently introduced a new feature to more accurately elucidate the association between genetic variation, functionality and disease. This new concept of position classes characterizes protein sequence positions by the distribution of variant effects on protein function. We could show that these classes, toggles (binary, on/off behavior) and rheostats (gradient range of effects) affect computational effect predictors with a clear pattern. We further developed a machine learning approach to predict these classes solely based on a set of sequence based features. This is the base for a new and improved computational variant effect prediction, able to overcome key limitations currently present in virtually every computational prediction method. Yannick Mahlich Department of Biochemistry and Microbiology HFSP: High speed homology-driven function annotation of proteins Abstract: HFSP: High speed homology-driven function annotation of proteins The rapid drop in sequencing costs has produced many more (predicted) protein sequences than can feasibly be functionally annotated with wet-lab experiments. I will describe HFSP, a novel computation technique that uses a high-speed alignment algorithm, MMseqs2, to infer functional similarity of proteins on the basis of their alignment length and sequence identity. We demonstrate that defining functional similarity between any two proteins is only feasible up to a certain level, however inferring function by detecting the protein with highest similarity results in excellent predictions even at the most precise EC level. Additionally, we applied HFSP to existing functional annotations available Swiss-Prot to highlight inconsistencies in reported enzymatic activity. HFSP: High speed homology-driven function annotation of proteins The rapid drop in sequencing costs has produced many more (predicted) protein sequences than can feasibly be functionally annotated with wet-lab experiments. I will describe HFSP, a novel computation technique that uses a high-speed alignment algorithm, MMseqs2, to infer functional similarity of proteins on the basis of their alignment length and sequence identity. We demonstrate that defining functional similarity between any two proteins is only feasible up to a certain level, however inferring function by detecting the protein with highest similarity results in excellent predictions even at the most precise EC level. Additionally, we applied HFSP to existing functional annotations available Swiss-Prot to highlight inconsistencies in reported enzymatic activity.