KaZam: An integrated inference engine for assembly
Abstract
Cancer signaling is an example of a complicated system where interactions have important causal effects. Creating mechanistic models, rather than correlative models, helps with understanding such systems. The goal of the Big Mechanism is to push forward tool development to support 1) the automatic generation of these models from natural language describing cancer signaling, 2) the analysis of these models to improve understanding, and 3) the generalization of our methods and tools for new datasets and new problems. The goal of our work for the remaining parts of the Big mechanism project is to redesign and redevelop our existing inference engine Syndra, which was designed and developed for biological signaling processes, to be able to integrate semantic, ontological reasoning with reasoning about the underlying properties for food security problems, and to develop analysis methods for the causality detection. The plan will have the following measurable milestones:Ê 1. Reasoning about domain knowledge, subsumption, and refinement at the level of semantics and ontologies. Syndra is a reasoning tool that supports semantic reasoning. At a semantic level, Syndra supports reasoning about how the statement ÒMEK is phosphorylated at site S222Ó implies the statement ÒMEK is serine-phosphorylated,Ó which implies the statement ÒMEK is phosphorylated,Ó and how all of these imply the statement ÒMEK is active.Ó In Syndra, typical biological actions, such as activation, phosphorylation, binding, and dephosphorylation, are defined as first-order logical predicates, and first-order logical formulae are used to encode these English statements. Considering the statements used and the discrete models generated for the food security problems, we need to firstly design a new translator to encode English statements in food security related papers into 3-valued logical formulae. The 3-valued logic is an extension to the propositional logic where each variable can be assigned into -1, 0, and 1. Then, we need to replace the Z3 SMT solver by an SAT solver, which is more suitable for the logical solving for the 3-valued logic.Ê 2. Scaling semantic reasoning. We should ensure that semantic reasoning scales in our system before moving on. The allowed semantic reasoning in Syndra is limited in statements describing the Ras-associated pathway. Efforts need to be put into releasing this limitation by adding more types of relations mentioned in food security papers. The ultimate goal is to help with the assembly of discrete models from food security related statements in a sound way. 3. Integrated reasoning with the structure of causal models. After redesigning and redeveloping a tool that supports the assembly of discrete models using logical inference, the next step is to integrate it with reasoning about the underlying causations leading to a certain situation. We might not just want to know that food insecurity will drop below a critical level in a certain region within 3 months, for instance, but that an event or a sequence of events we derive can lead to such a situation. We plan to study such causalities by considering both the Bayesian causality and the counterfactual manner of causation. In detail, for the Bayesian causality, we will encode a large enough set of executions leading to a given situation into time series, and develop model search algorithms to detect key causal events. For the counterfactual causality, in order to figure out whether the event p is the counterfactual cause to the blocking of event q, we need to check whether, with a really high probability, we have if event p had not happened, event q would have. To achieve this, restimulation for individual event traces will be considered. With strong mathematical guarantees, we will generalize our methods and tools to solve important open problems for assembling mechanistic models from natural language and study the corresponding underlying causalities for food security problems.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Jul 24, 2019
- Source ID
- W911NF1710073
Entities
People
- Qinsi Wang
Organizations
- Army Contracting Command
- Massachusetts Institute of Technology
- Office of the Secretary of Defense