KoBIS November 2018
06.11.2018 um 16:15 in Raum C10.3.36, THM Gießen
Christian Thomae Viegas, Fachbereich MNI, Technische Hochschule Mittelhessen
"Prediction of antiviral peptides with various machine learning algorithms"
Most viral infections are still extremly difficult to treat. Many existing drugs only relief the symptoms, but offer no cure. Antiviral peptides (AVP) could be used prophylactically before infection or afterwards. They are very common in nature, working as a first line of defence in many organisms. Synthesis and assessment of antiviral action on the other hand are very costly and time-consuming in a laboratory environment. Well trained machine learning algorithms can rapidly predict AVPs from sequence information. In combination with exponentially growing, dedicated sequence-databases an in silico identification of potential AVPs can be established. Promising candidates can then be subjected to further testing in a wet-lab environment. At first, indicative features characterising AVPs have to be selected, based on classification properties. Those can be from the chemical, physical or biological domain. Different methods will be applied to reduce the high dimensional featurespace to a smaller feature-set covering most of the relevant information about AVPs. This is a crucial step, as these features are used for training and define the classifiers. A pipeline of different machine learning approaches will be set up next, combining support vector machines, artificial neural networks and random forest. Furthermore, hyperparameters steering the classification algortithms will also be subjected to optimasation. A unified score weighting the outcome of different algorithms is used as an overall prediction-score. For that matter all subtleties of the different approaches will have to be considered. Finally, extensive laboratory validation will be conducted with a random representative set of predicted AVPs and peptides, predicted to have no (antiviral) effect.
Fabian Tann, Fachbereich LSE, Technische Hochschule Mittelhessen