Location: Collegezaal C
Setup: two research talks, and drinks afterwards, sponsored by Delft Data Science.
Assistant Prof. Sicco Verwer (TU Delft) works on machine learning with applications to cyber security and software engineering.
Abstract: In this talk I describe key algorithms for learning state machines from trace data. I discuss their theoretical guarantees, the tools that implement them, and how we use them in practice. Essentially, these algorithms aim to turn a black-box, such as a software component, into a white-box model, in this case a state machine. Important applications of this technology is to automatically reverse engineer network protocols, interfaces, or legacy software. In cyber security, we use is to learn models recognising malware behaviour and to detect attacks in SCADA systems.
Assistant Prof. Joana Gonçalves (TU Delft) will talk about algorithms to reveal mechanisms of diseases.
Abstract: The molecular characterization of patient or disease model samples offers crucial insight into the molecular basis of disease, enabling the identification of diagnostic and prognostic biomarkers and the discovery of promising new targets for treatment. Diagnostic biomarkers used in the clinic are typically associated with the disruption of normal biological function in disease. They are often detected as changes in gene expression (mRNA) and/or harmful variations in the genome (DNA) by comparative analysis of disease versus healthy samples. However, identifying prognostic biomarkers and potential targets for therapy requires additional mechanistic knowledge, and the inherently adaptive biological processes such as disease progression and drug response are better studied along the temporal dimension.
In this talk, I will discuss how we use string matching techniques to develop efficient multi-way clustering algorithms for time course data. These enable the grouping of genes exhibiting coordinated behavior over time frames that may not span the entire timespan of an experiment (local patterns). Our methods allow for partially overlapping groups and potentially delayed temporal patterns to capture well-known biological phenomena. I will go over some of our recent work combining (i) multi-way clustering on temporal gene expression profiles with (ii) network-based diffusion on interactions between regulators and target genes to identify pathways and regulators driving important biological processes in prostate cancer. We experimentally validated effects of two identified regulators in collaboration with researchers at the Netherlands Cancer Institute, which will be investigated further. Finally, I will outline current challenges and future research directions motivated by technological developments in the field.