His core competences lie in 2D graph based representations providing a good trade-off between expressiveness and computational complexity. This combines methods such as graph mining, supervised feature selection, dimensionality reduction, as well as activity-specific similarity indices. The current Lazar prediction system offers automatic applicability domain estimation and full details for every single prediction to expert users from the biochemical domain in a user friendly web interface.

Solutions in the biochemical domain are appealing both for biochemistry and computer science. Reliable and user friendly computer based models and alerts in drug discovery and risk management, particularly for early stages of development, are high priority for both regulators and the pharmaceutical industry, especially with regard to endpoints which are not fully elucidated so far. On the other hand, this constitutes a challenging field for computer scientists. There are many possible representations for chemicals that lend themselves more or less to data mining tasks. Creating an inductive database therefore implies solutions to sophisticated problems such as graph mining, pattern recognition, and multivariate modeling with dimensionality reduction, also relevant to other areas of applied computer science.

Besides mining and modeling I am concerned with structural alerts, i.e. fragments of chemical structures associated with biological activity, mostly on a cellular level. The objective is to establish an inductive database of compounds and activity values that users can query for fragments relevant to the endpoint of interest, providing, among others, frequency and statistical constraints. For this purpose efficient subgraph enumeration including pruning techniques for statistical and frequency constraints are employed.

Publications:

• Journals and Book Chapters:
  • C. Helma, A. Maunz: "Prediction of toxic effects of pharmaceutical agents" in "Pharmaceutical Data Mining: Approaches and Applications for Drug Discovery", edited by Konstantin V. Balakin, M.Sc., Ph.D., D.Sc., Wiley 2009 (Wiley, Amazon). [abstract]
  • A. Maunz, C. Helma: "Prediction of chemical toxicity with local support vector regression and activity-specific kernels" in SAR and QSAR in Environmental Research, Volume 19, Issue 5 and 6 July 2008 , pages 413 - 431. [pdf, bib]
  • F. Hammann, H. Gutmann, U. Jecklin, A. Maunz, C. Helma, J. Drewe: "Development of Decision Tree Models for Substrates, Inhibitors, and Inducers of P-Glycoprotein" in Current Drug Metabolism, Volume 10, Number 4, May 2009 , pp. 339-346(8). [bib]
• Conferences:
  • A. Maunz: "Instance-based Regression Models for Quantitative Biological Activities using Support Vector Machines and Multilinear Models", presented at the Scarlet Workshop on in silico methods for carcinogenicity and mutagenicity, Milano, April 2008. [pdf]
  • A. Maunz, C. Helma: "New Lazar Developments", presented at the eCheminfo Community of Practice InterAction Meeting, Autumn 2008, Bryn Mawr College, Philadelphia (13-17 October 2008). [ppt, pdf]
  • A. Maunz, C. Helma, and S. Kramer: "Large Scale Graph Mining using Backbone Refinement Classes". In KDD '09: Proceedings of the 15th ACM SIGKDD international conference on Knowledge discovery and data mining. [pdf, bib]

    © ACM, 2009. This is the author's version of the work. It is posted here by permission of ACM for your personal use. Not for redistribution. The definitive version was published in KDD '09: Proceedings of the 15th ACM SIGKDD international conference on Knowledge discovery and data mining, pages 617-626, 2009, http://doi.acm.org/10.1145/1557019.1557089.

    Video lecture: slides 1-3.exe, 4-6.exe, 7-10.exe, 11-15.exe, 16-19.exe, 20-23.exe or view the conference recordings.
  • A. Maunz, C. Helma, and S. Kramer: "Large Scale Graph Mining using Backbone Refinement Classes". In 7th International Workshop on Mining and Learning with Graphs (MLG 2009). [pdf, pptx]

Unpublished Material:

• Related to Publications:
  • A. Maunz: "On the Number of Backbone Refinement Classes", derivation and proof of a formula for counting the number of BBRCs in a perfect binary tree. Comparison to the total number of subtrees [pdf bbrc-no.pdf].
  • A. Maunz: "On the Co-Occurrence and Diversity of Backbone Refinement Classes", euclidean embedding of BBRC features and instances in 2D as well as feature similarity alike to ORIGAMI approach [pdf bbrc-rep.pdf].
  • A. Maunz: "Support Vectors and the Margin in a Nutshell", brief outline of support vector theory and the concept of margin. (see ch. 1,2 and 7 of Schölkopf and Smola, 2002) [pdf sv-margin.pdf].

Links:

• Related to Publications:
  • "Graph Mining and Graph Kernels", video lecture by Karsten Borgwardt and Xifeng Yan in KDD '08 [html].
• Other material:
  • "Virtuelle Versuchskaninchen", Feature in Deutschlandradio about Lazar [html].
  • "Mehr Tierversuche durch Chemikalienrichtlinie Reach", TV report in 3sat about animal testing and alternative test methods in the EU, featuring an interview with Christoph. [html].

Software:

• Large-Scale Graph Mining using Backbone Refinement Classes, presented at KDD '09.

  An approach to large-scale graph mining based on so-called backbone refinement classes. The method efficiently mines tree-shaped subgraph descriptors under minimum frequency and significance constraints, using classes of fragments to reduce feature set size and running times. The classes are defined in terms of fragments sharing a common backbone. The method is able to optimize structural inter-feature entropy as opposed to occurrences, which is characteristic for open or closed fragment mining.

  View a flash-animated 2-D embedding of backbone refinement class features and database instances.

  Links to the datasets used in this study:

  • CPDB carcinogenicity and mutagenicity datasets
    To reproduce experiments (Fminer arguments) : Frequency 6 (-f6); Use aromatic perception (-a), as described in the paper.
  • Yeast anti-cancer datasets
    To reproduce experiments (Fminer arguments) : Frequency 100 (-f100); Use aromatic perception (-a), as described in the paper.

Short CV:

• 2004-2005: Cooperation at the Lehrstuhl für Maschinelles Lernen und Natürlichsprachliche Systeme as "Wissenschaftliche Hilfskraft"
• Since 2005: Cooperation with the sens-it-iv project as "Wissenschaftliche Hilfskraft"
• July 2007: Obtained the degree of Diplom-Informatiker
• Since July 2007: Scientific position at FDM, Uni Freiburg.
• Since September 2008: Ph.D. student at TU Munich, Bioinformatics.

Biography:

I have always been fascinated by computer programs. At age eleven, I spent all my money for a four year old Atari ST computer. This early connection to formal systems inspired me to learn more about their potentials and limitations, and soon encouraged me to work precisely and carefully, and to obey to correctness and completeness.

After high school with emphasis on mathematics and english I decided not to attend university immediately. Instead, I completed a 3-year professional training at Deutsche Telekom AG (Brands are T-Com, T-Mobile, etc.). Working in several departments, I learned to satisfy the needs of business customers and assisted in setting up the ADSL network in the area.

After successful completion I started my degree in computer science at Freiburg University in October 2001. By that time I had gained enough experience to know that it would be challenging and loaded with applied mathematics. The foundation courses turned out to be the most ambitious part. For the main courses I decided to move towards data mining, since I have always been fascinated by solving hard practical problems.

In 2004, positions for scientific student workers were offered at the Machine Learning and Natural Language Processing Lab (Prof. Luc de Raedt). I applied and was offered the opportunity to work for Dr. Christoph Helma and Dr. Andreas Karwath. Since they were concerned with feature and string mining, respectively, I was assigned tasks such as creating a molecule viewer Java applet for the Lazar web interface (C. Helma. Lazy structure-activity relationships (lazar) for the prediction of rodent carcinogenicity and Salmonella mutagenicity. Molecular Diversity, 2006. [pdf]) and optimizing the string miner FAVST for buffering, necessary for large string databases, such as DNA sequences (S. D. Lee and L. De Raedt. An Efficient Algorithm for Mining String Databases Under Constraints. Lecture Notes in Computer Science 2004 [pdf]).

In July 2007, I received my diploma in computer science (German equivalent to M.Sc.) for my quantitative extension to the Lazar algorithm. Consecutively, I was offered a position at Freiburg Center for Data Analysis and Modeling (Prof. Jens Timmer) in the Christoph Helma working group.In September 2008, I joined TU Munich as a Ph.D. student.