Most of our work is applied AI/ML, including Deep Learning, but we have developed novel methods too, including:
Active Learning and Sequential Model Based Optimization
Most of the algorithmic work in my group is now focused on the development of methods to automate scientific discovery and engineering. To do this, we employ Active Learning and Sequential Model Based Optimization. Both techniques are concerned with maximizing the return on investment in an online setting.
Primary publication in this area
Asynchronous Parallel Bayesian Optimization for AI-driven Cloud Laboratories, T. Frisby, Z. Gong, C.J. Langmead ISMB/ECCB; Bioinformatics, 2021
Bayesian Statistical Model Checking & Temporal Logic-based inference in Graphical Models
Model Checking is a technique for formally proving whether finite-state system satisfies a given specification written in temporal logic. Historically, Model Checking has been used to verify the properties of computer hardware and software. My group introduced the idea of performing inference in graphical models (specifically, Dynamic Bayesian Networks) against Temporal Logic Specifications, we also introduced the first Bayesian approach to Model Checking stochastic systems.
Primary publications in this area
Generalized queries and Bayesian statistical model checking in Dynamic Bayesian Networks: Application to personalized medicine, C.J. Langmead, Proc. 8th Int. Conf. on Computational Systems Bioinformatics (CSB) pp. 201-212, August, 2009
A bayesian approach to model checking biological systems, SK Jha, EM Clarke, CJ Langmead, A Legay, A Platzer, P Zuliani International conference on computational methods in systems biology, 218-234, Sept. 2009
Parameter Synthesis for Dynamical Systems Models
Machine Learning usually involves estimating the parameters of a given model by fitting to data. However, it is also possible to synthesize parameters against temporal logic formulas. That is, finding parameter combinations that ensure that the model exhibits a given behavior. My group has developed and applied a variety of parameter synthesis methods for dynamical systems models, often for models of biological systems. Some methods can also be used to design controllers for closed-loop systems
Primary publications in this area
Symbolic Approaches to Finding Control Strategies in Boolean Networks, SKJ C.J. Langmead, Journal of Bioinformatics and Computational Biology 7 (2), 323-338
Parameter synthesis in nonlinear dynamical systems: Application to systems biology A Donzé, G Clermont, CJ Langmead Journal of Computational Biology 17 (3), 325-336, 2010
Synthesis and infeasibility analysis for stochastic models of biochemical systems using statistical model checking and abstraction refinement, SK Jha, CJ Langmead, Theoretical Computer Science 412 (21), 2162-2187, 2011
Synthesis of insulin pump controllers from safety specifications using bayesian model validation SK Jha, RG Dutta, CJ Langmead, S Jha, E Sassano Proceedings of the Tenth Asia Pacific Bioinformatics Conference
Stochastic computational model parameter synthesis system, SK Jha, CJ Langmead, US Patent 9,558,300 (2017)
Game Theoretic models of Molecular Evolution & Drug Resistance
My group introduced the idea of modeling the response of complex distributions over biological sequences (ex. viral quasispecies or genetically heterogeneous tumors) to selective pressures (i.e., drugs) as a multiplayer game, where the residue positions of the virus/tumor protein comprise one 'team', and the constituents of a combination therapy comprise the other 'team'. Given this, we showed that it is possible to compute the correlated equilibrium of the game, which represents the optimal joint strategy for both teams. Such solutions can be used to either (a) predict which resistance mutations will arise, to a given drug (or drug cocktail), or (b) design optimal drug cocktails. The specific formalism used to compute the strategy is a graphical game.
Primary publication in this area
Approximating correlated equilibria using relaxations on the marginal polytope H Kamisetty, EP Xing, CJ Langmead, ICML, 2011
Studying Rare Events in Stochastic Systems
When studying biological systems, it is often the rare events/behaviors that are the most interesting and consequential. Unfortunately, studying those events is difficult, precisely because the events are rare. This remains true even if one uses computational models of the system. To address this challenges, we have developed methods to accelerate discovery of interesting behaviors.
Primary publications in this area
Exploring Behaviors of Stochastic Differential Equation Models of Biological Systems using Change of Measures SK Jha, CJ Langmead Computational Advances in Bio and Medical Sciences (ICCABS), 2011
Decision procedure based discovery of rare behaviors in stochastic differential equation models of biological systems, AK Ghosh, F Hussain, SK Jha, CJ Langmead, S Jha 2012 IEEE 2nd International Conference on Computational Advances in Bio and Medical Sciences (ICCABS), 2012
Discovering rare behaviours in stochastic differential equations using decision procedures: applications to a minimal cell cycle model AK Ghosh, F Hussain, S Jha, CJ Langmead, SK Jha International Journal of Bioinformatics Research and Applications 2 10 (4-5), 2014