Methods for analyzing molecular data are in rapid progress, but still are far from catching up with high-throughput experimental data collection techniques. Our work is focused on closing the gap between the amassing wealth of proteomic and genomic information and the models that aid understanding of the data. In our work, we aim at incorporating prior knowledge about biology of complex processes and diseases into model creation. In this way, the models are not created de novo, by using only the information from a single experiment. We work on novel machine learning methods for creating such models. Currently, we apply our new methods to proteomic measurements relating to impaired wound healing, and to multi-modal (genetic, epigenetic, expression, proteomic) studies of cancer.

Complex networks gained recognition in recent years as a universal framework for modeling many natural phenomena. In molecular biology, they are convenient tool for representing functional, structural or causal interactions between entities in cellular networks. We focus on metabolic network, analyzing the conservation and differences in the robustness to attacks for networks of closely related species and comparing them with networks of unrelated organisms.

Computational and statistical methods are at the core of efforts aimed at understanding the paths of protein evolution. We study what general rules underlie the divergence of protein structure as its genetic sequence changes.

The ensemble methods form one the leading approaches to machine learning, with algorithms like bagging or AdaBoost boosting achieving state-of-the-art accuracy in a wide range of applications. These methods are based on the idea to enhance the classification or clustering ability of the underlying learning method by constructing a number of individual classifiers or clusters. Then, their answer is combined, e.g. by a weighted average, to obtain the final decision. Most of the ensemble methods are proposed for the supervised learning scheme. However, recently, applications of ensemble scheme to unsupervised learning are emerging. In our research, we focus on developing new classification methods based on boosting. We also work on ensemble feature selection methods and diversity measures of ensembles.