Presentation on theme: "Nmrbox.org NMRbox: TRD 3 A probabilistic core as a coherent inference engine PINE+ Core Extend functionality through the new core PINE+: Assignment, use."— Presentation transcript:
nmrbox.org NMRbox: TRD 3 A probabilistic core as a coherent inference engine PINE+ Core Extend functionality through the new core PINE+: Assignment, use of structure data, RNA Reproducibility – “revisable executable documentation” Side-by-side, reproducible, probabilistic and deterministic
nmrbox.org Why Bayes? argmin E total = E empirical + λE experimental Bayes: Extremizes the log-likelihood. Consistent & extensible determination of parameters Constrain: NOEs not distances. J’s not dihedrals etc. Consistency + Extensibility Reproducibility
nmrbox.org What is PINE and PINE Core? PINE is the probabilistic Bayesian engine for Data verification (LACS) Secondary structure determination (PECAN) Assignment (PISTACHIO) Peak picking (HIFI) Bayesian updating (PINE) Bayesian updating + deterministic constraints (ADAPT-NMR) RNA assignment (RNA-Pairs) LACSPECAN PISTACHIO HI FI LACS PECAN PISTACHIO LACS PECAN PISTACHIO HI FI ADAPT-NMR
nmrbox.org Corollary In our view, Pine has been a highly successful probabilistic model. The inference core has proven to be very powerful – an excellent setting for building a structure determination protocol for: Repeatability and Reproducibility Coexistence of probabilistic and deterministic paradigms Extensibility to support new functions through plugins Additionally empowering: Revisable executable documentation Document as you go, execute any time, reuse.
nmrbox.org PINE+ PINE+ is the collection of applications Containing a core probabilistic engine The core will support all existing tools HIFI, PINE, PECAN, ADAPT-NMR, LACS, RNA-PAIRS And, New assignment tools NOESY, RNA, Structure-based PINE+ will support probabilistic and deterministic approaches PINE+ Will additionally enable reproducible results Playback of blocks or the entire project (repeatability)* Edit & Replay of blocks or the entire project (reproducible)* Template (reuse)
nmrbox.org Key Innovations The complete process of structure determination is captured and executed using one language, one environment, one paradigm, in open source form, using open technologies. As the user steps through structure determination steps, a living “executable document” is created – repeatability. “Blocks” of a structure determination “document” can be modified with new data or process – reproducibility. A “document” can be used as a “template”; replacing, or modifying blocks as needed – reuse.
nmrbox.org Impact Opportunity for Enabling repeatable and reproducible structure determination Empowering other packages with probabilistic functionality Supporting side-by side deterministic and probabilistic tools Creating end-to-end open source tools Built-in probabilistic functions for Peak picking, backbone assignment, side-chain assignments, secondary structure prediction, verification, NOESY assignment, RNA assignment, pipeline/network definition, scientific documentation, capture and replay, edit and execute.
nmrbox.org Progress to date Things under the hood Environment Protocol Engine Demonstration Interface Desktop PINE Executable document
nmrbox.org Progress in context
nmrbox.org Things under the hood Environment High performance, dynamic, distributed, parallel, scientific computing Protocol JSON for specification and exchange, Pandoc for data transformation Engine Inference-as-a-service using a resource-oriented model Simple ( 6 components ) relation, learner, predictor, transformer, attribute, schema
nmrbox.org Demo Show interface and interaction Start a PINE job Display output of PINE Show block idea Show repeatability idea
nmrbox.org Novel sampling and BP strategy Key computational advantage of PINE+ core