Our goal is to develop and analyze models of large biochemical networks that typically arise in the study of signalling pathways. Specifically, we have been addressing a basic problem that arises when one constructs ODEs based bio-pathways models; namely, one must first estimate the values of many unknown rate constants and initial concentrations using limited noisy data. We have developed decomposition methods using which a large model can be broken down into smaller models. One then does parameter estimation on the smaller models followed by reconciling conflicting estimates of shared parameters via the well known technique called belief propagation. We have also been building a powerful approximation technique through which a large ODEs based model (with many unknown parameters) is first converted into a probabilistic graphical model called a dynamic Bayesian network. All subsequent analysis including parameter estimation is then carried out on the approximated –much simpler- model with the help of standard Bayesian inferencing techniques. We have applied this method successfully in a number of biologically relevant settings. We are also continuing to extend its scope and applicability along multiple dimensions. In particular we are developing GPU-based implementations of all our algorithms as well as formal verification techniques such as probabilistic and statistical model checking.

Recently, we have started research on network inferencing. Our goal is to construct novel synthesis methods by which one can learn dynamic models (to start with, a dynamic Bayesian network) using time series data and prior knowledge networks.

We have ongoing collaborations with biologists involving the study of
(i) TLR3/TLR7 signaling pathways and their cross talks in the presence of multiple infections
(ii) DNA damage/response signalling under a range of oxidative stresses
(iii) Chromosome localizations during T–cell activations.