DARYA CHUDOVA


Home Department
Information and Computer Science

Thesis Advisor
Padhraic Smyth - Information and Computer Science

Co-thesis Advisor
G. Wesley Hatfield – Microbiology and Molecular Genetics, College of Medicine

Email


Darya Chudova is studying the problem of joint reconstruction of structure and dynamics of regulatory sub-networks in yeast from available experimental data. Smyth’s lab presents a family of probabilistic models and inference algorithms that attempt to simultaneously recover the structure of interactions between the nodes in the network and to quantify these interactions. She makes specific assumptions about the structure of interactions and the dynamics in order to constrain the set of probable models. The models assume a small number of tightly interacting "master" genes in the core of the network that govern the dynamics of the system. The rest of the genes follow the dynamics of their "master" nodes with some time lag, without influencing the core dynamics. Given the structure of the network, the dynamics is described by a system of differential equations that can be modeled with a neural network. She uses Bayesian approach and allow to incorporate prior knowledge about (1) the out-degrees of the "master" genes, (2) specific interactions that take place (known pathways) and (3) general structure of the regulation matrix (known co-regulated families of genes etc.). The approach extends commonly used techniques that de-couple structure from dynamics, and perform clustering of expression profiles, followed by modeling the dynamics within the learned clusters.

The computational challenge of the problem is in simultaneous recovery of structure of the interactions (discrete optimization problem) and dynamics of the system (continuous optimization problem). She, in collaboration with Dr. Mjolsness’ lab developed the Gibbs sampler to address this problem, and demonstrated that it can successfully recover the data generating models on synthetic data sets with characteristics similar to that of real gene expression data. At present, are working on gathering applicable prior information and applying the models to the gene expression time course data in yeast. She plans to incorporate the binding ratio data and protein/protein interaction data as a future step.

Publications

D. Chudova, P. Smyth, Anaylsis of pattern discovery in sequences using Bayes error rate framework. (Acccepted for publication in the Data Mining and Kowledge Discovery journal.)

N. Kh. Rozov, V. G. Sushko, D. I. Chudova. Differential equations with degenerate coefficient by order derivative. Fundamentalnaya I prikladnaya matematika (Fundamental and applied mathematics), 1998. v. 4, # 3, pp.1063-1095 (In Russian)

D. Speights, J. Brodsky and D. Chudova. Using Neural Networks to Predict Claim Duration in the Presence of Right Censoring and Covariates. Casual Actuarial Society Forum, Winter 1999

Conference Proceedings

Probabilistic Models for Joint Clustering and Time-Warping of Multidimensional Curves. D. Chudova, S. Gaffney, P. Smyth. Submitted to UAI-03 (Uncertainty in Artificial Intelligence) conference.

Translation-invariant clustering of multidimensional curves. D. Chudova, S. Gaffney, E. Mjolsness, P. Smyth, Submitted to ACM SIGKDD-03 conference.

Pattern discovery in sequences under a Markov assumption. D. Chudova and P. Smyth. Proceedings of the Eighth ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, Edmonton, 2002.
Unsupervised identification of sequential patterns under a Markov assumption. D. Chudova and P. Smyth. Presented at the ACM SIGKDD 2001 Workshop on Temporal Data Mining, San Francisco, 2001.

Towards scalable support vector machines using squashing. D.Pavlov, D. Chudova and P. Smyth. Proceedings of the Sixths ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, Boston, 2000

Benchmarking of Different Modifications of the Cascade Correlation Algorithm. D. I. Chudova, S. A. Dolenko, Yu. V. Orlov, D. Yu. Pavlov, I. G. Persiantsev. Proc. 3rd International Conference on Adaptive Computing in Design and Manufacture, 1998, pp.339-344.

Development of a Statistics Based System for Fluorescent Diagnostics of Organic Pollution in Water. Yu.V.Orlov, I.G.Persiantsev, D.I.Chudova, D.Yu.Pavlov, S.M.Babichenko. Proc. of the 3rd EARSeL Workshop on Lidar Remote Sensing of Land and Sea, 1997, pp.157-162.