System and method for integrating data for precision medicine
Assigneedatma, Inc.
InventorsGanapati Srinivasa, Melvin Lathara, Jaclyn Smith, Nalini Ganapati, Hollis Wright, Kavya Kannan
Life of this patent
Estimating the term…
Abstract
Methods and systems are provided for a platform and language agnostic method for generating inter- and intra-data type aggregations of heterogeneous disparate data upon which various operations can be performed without altering the structure of the query or resulting distributed data set representation to account for which specific data sources are included in the query.
Claims19
A method for generating a causal model with a distributed data integration system, the method being implemented by a manager adapted to service requests from a client device and comprising one or more processors executing computer program instructions that, when executed, perform the method, the method comprising:
- receiving, at the manager, a causal model construction job from the client device via a job definition service requesting construction of the causal model using one or more patient cohorts;
- distributing, to each computational node and without exposing underlying data, at least one task associated with construction of the causal model, wherein the causal model construction job includes a plurality of parameters defining the causal model construction job, the plurality of parameters including model parameters, the one or more patient cohorts, and expert knowledge;
- receiving, at the manager, a first update to the causal model from a first computational node of a plurality of computational nodes, each computational node configured to process biomedical data from patients specified by the one or more patient cohorts in order to generate updates to the causal model;
- distributing, with the manager, the first update to the causal model to remaining computational nodes of the plurality of computational nodes;
- receiving, at the manager, one or more additional updates to the causal model from one or more of the remaining computational nodes;
- determining that the first update and the one or more additional updates have caused the causal model from each computational node to converge to a final causal model;
- and
- returning the final causal model to the client device.
A method for generating a causal model with a HIPPA compliant distributed data integration system, the method being implemented by a computational node comprising one or more processors executing computer program instructions that, when executed, perform the method, the method comprising:
- receiving a causal model construction job defining parameters for construction of the causal model using one or more patient cohorts without exposing data;
- responsive to receiving the causal model construction job, sending a definition for each of the one or more patient cohorts to a cohort service, the cohort service configured to extract biomedical data from one or more data sources based on the definition for each of the one or more patient cohorts;
- processing the biomedical data associated with the one or more patient cohorts to generate a first update to the causal model, the one or more patient cohorts and associated biomedical data identified based on the causal model construction job;
- sending the first update to a manager in communication with a plurality of additional computational nodes;
- receiving, from the manager, one or more additional updates to the causal model made by one or more of the plurality of additional computational nodes;
- iteratively updating the causal model based on the one or more additional updates and/or additional processing of the biomedical data until a final updated causal model is created;
- determining that the final updated causal model has a fitness metric above a threshold;
- and
- returning the final updated causal model to the manager.
A method for generating a causal model with a HIPPA compliant distributed data integration system, the method being implemented by or more processors executing computer program instructions that, when executed, perform the method, the method comprising:
- receiving, at a manager adapted to service requests from a client device, a causal model construction job, the causal model construction job including one or more model parameters, one or more patient cohort definitions, and expert knowledge to be used in construction of the causal model;
- mapping, with a cohort service, each patient cohort definition with respect to a data integration schema including integrity constraints to form, for each patient cohort definition, one or more query segments;
- extracting, with the cohort service and while maintaining patient privacy, biomedical data of a plurality of patients from a plurality of siloed data sources according to the one or more query segments and storing the biomedical data as a plurality of data frames;
- distributing, with a job definition service and/or the manager, the causal model construction job to a plurality of computational nodes, each computational node configured to process data from one or more of the plurality of data frames to generate updates to the causal model;
- receiving, at the manager, a first update to the causal model from a first computational node of the plurality of computational nodes;
- distributing, with the manager, the first update to the causal model to remaining computational nodes of the plurality of computational nodes;
- receiving, at the manager, one or more additional updates to the causal model from one or more of the remaining computational nodes;
- determining, at the manager, that the first update and the one or more additional updates have caused the causal model from each computational node to converge to a final causal model;
- and
- returning the final causal model to the client device.
Dependencies are inferred from the claim text.