DTAI

Monday 23 May 2011, at 16h30 in Celestijnenlaan 200A Auditorium 00.225

Peptide Identification using Mass Spectrometry Data
By Eduardo De Paula Costa (PhD student DTAI)

Peptides are short proteins that play a key role in many physiological processes such as blood pressure regulation, water balance and glucose metabolism. Hence, peptide identification is an important step in understanding how these processes work. Currently, one of the most important techniques for peptide identification is Mass spectrometry (MS), an analytical technique used for measuring the molecular mass of a biological sample. MS produces various types of data. The most common data representation is the mass spectrum. This spectrum is a plot containing the chemical analysis of a molecule (a peptide, for example) and it is used for the molecule identification. As spectrum identification can be a complex and time consuming task, many computational methods and tools have been proposed to assist researches in this task. In this presentation, I will introduce the main aspects involving peptide identification using mass spectrometry data and I will discuss the main computational approaches currently used to interpret mass spectra. More specifically, I will present a method that we are currently developing, which extends the data base search approach for peptide identification by considering additional genomic information during the spectrum identification task.

More info

Monday 9 May 2011, at 16h30 in Celestijnenlaan 200A
Auditorium 00.225

First-order Bayes-Ball
By Nima Taghipour (PhD student DTAI)

Probabilistic logic models bring the expressive power of first-order logic to probabilistic models, enabling them to capture both the relational structure and the uncertainty present in data. Inefficient inference, however, is a bottleneck in these models, affecting also the cost of learning these models from data. One issue that increases the cost of inference is the presence of irrelevant random variables. The Bayes-ball algorithm can identify the requisite variables in a propositional Bayesian network and thus achieve more efficient inference by ignoring irrelevant variables. We presents a lifted version of Bayes-ball, which works directly on the first-order level, and show how this algorithm applies to (lifted) inference in directed first-order probabilistic models.

On Thursday September 3, the Declarative Languages and Artificial Intelligence research group of the Katholieke Universiteit Leuven organizes a symposium on "Recent Trends in Machine Learning".

After the symposium, Tom Croonenborghs will defend his PhD thesis entitled "Model-Assisted Approaches for Relational Reinforcement Learning".

Attending the symposium and/or PhD defense is free, but registration is required (by August 31).

More info