Causal inference-based investigation of antimicrobial resistance

1 . A method to perform causal analysis on data relating to Anti-Microbial Resistant (AMR) bacteria, comprising:

receiving patient data, clinical trial data, and bacterial kinetic model data, wherein the patient data relates to instances of AMR infections;

converting the patient data, clinical trial data, and bacterial kinetic model data into normalized attribute vectors having a normalized attribute vector form, the form being a vector format including at least a model portion that contains model attributes describing aspects of a bacterial kinetic model, an experiment portion that contains experiment attributes describing an experiment in a clinical trial, and a data portion that contains data attributes providing values representing the patient data and experimental results of the clinical trial;

loading the normalized attribute vectors into a clinical trial data store that contains data from multiple heterogeneous clinical trials, all in the normalized attribute vector formal;

creating, in a graph database, a graph database instance for a set of the normalized attribute vectors stored in the data store, including:

adding, to the graph database instance, respective nodes corresponding to the normalized attribute vectors of the set; and

for each of the normalized attribute vectors:

comparing the model attributes with the model attributes of normalized attribute vectors corresponding to other nodes in the graph database instance using a first similarity score;

creating respective edges in the graph database instance between nodes corresponding to compared normalized attribute vectors having a first similarity score within a first predetermined threshold;

comparing the experiment attributes with the experiment attributes of normalized attribute vectors corresponding to the other nodes in the graph database instance using a second similarity score; and

creating respective edges in the graph database instance between nodes corresponding to compared normalized attribute vectors having a second similarity score within a second predetermined threshold;

creating, in a category database, a category database instance for the set of normalized attribute vectors, including, for each of the normalized attribute vectors:

comparing the model attributes to model attributes associated with categories of other normalized attribute vectors in the category database instance; and

associating, in the category database instance, the normalized attribute vector with categories corresponding to model attributes of the other normalized attribute vectors in the category database instance that match the model attributes of the normalized attribute vector;

creating a causal inference model from data in the clinical trial data store, the graph database instance, and the category database instance corresponding to the normalized attribute vectors; and

performing a causal analysis on the causal inference model.

2 . The method of claim 1 , wherein the receiving, converting, and loading are performed by a loader making use of a multi-format combinatorial parser.

3 . The method of claim 2 , wherein the patient data includes response curve data for a particular drug, the bacterial kinetic model data includes data about biological markers not necessarily related to AMR and time limit threshold data for the biological markers, and wherein the causal inference model is to determine whether the particular drug not currently identified as effective against AMR bacteria, is effective against AMR bacteria.

4 . The method of claim 2 , wherein the bacterial kinetic model data includes data about AMR mechanisms of resistance for a set of bacteria, and the clinical trial data includes vectors of attack on the set of bacteria, efficacy of the vectors of attack of a particular drug on the set of bacteria, and wherein the causal inference model is to determine whether the particular drug bypasses the AMR mechanisms of resistance of the set of bacteria and is effective on the set of bacteria.

5 . The method of claim 2 , wherein the bacterial kinetic model data includes data about a bacteria's mechanisms of resistance and the bacteria's vulnerabilities, the clinical trial data includes data of a first drug's efficacy against the bacteria and data of a second drug's efficacy against the bacteria, and wherein the causal inference model is to determine whether the first drug and the second drug should be combined into a cocktail where either the first drug or second drug will bypass the bacteria's mechanism of resistance.

6 . The method claim 1 , further comprising:

receiving a query to search for causal inferences on data in the clinical trial data store,

wherein the causal analysis is performed on the created-data causal inference model in response to receiving the query.

7 . The method of claim 2 , wherein the multi-format combinatorial parser makes use of a machine learning model.

8 . The method of claim 7 , wherein the machine learning model is a large language model.

9 . The method of claim 1 , further comprising:

creating an empty normalized attribute form record, populating the created record with patient data, backfilling the created record with at least some data from the clinical trial data for a particular drug and at least some kinetic model data of the particular drug, and loading the created and populated record into the clinical trial data store.

10 . The method of claim 9 , further comprising:

checking the consistency of the created record, including:

computing the response of a patient to the particular drug based on the patient data;

comparing the computed response with a response predicted by the kinetic model data of the particular drug; and

if the computed response is within a predetermined threshold of the predicted response, indicating that the created record is verified and can be loaded into the clinical trial data store.

11 . The method of claim 10 , wherein the predetermined threshold is a predetermined standard deviation.

12 . The method of claim 1 , further comprising:

creating, in the category database instance, a new category associated with the model attributes of the normalized attribute vector whose attributes fail to match the model attributes of the other normalized attribute vectors in the category database instance.

13 . A system, comprising:

one or more processors; and

memory including at least one computer-readable medium storing instructions that, if executed by the one or more processors, cause the one or more processors to perform causal analysis on data relating to Anti-Microbial Resistant (AMR) bacteria, the causal analysis including operations comprising:

receiving in-field patient data, clinical trial data, and bacterial kinetic model data, wherein the in-field patient data relates to instances of AMR infections;

converting the in-field patient data, clinical trial data, and bacterial kinetic model data into normalized attribute vectors having a normalized attribute vector form, the normalized attribute form being a vector format including at least a model portion that contains model attributes describing aspects of a bacterial kinetic model, an experiment portion that contains experiment attributes describing an experiment in a clinical trial, and a data portion that contains data attributes providing values representing the patient data and experimental results of the clinical trial;

loading the normalized attribute vectors into a clinical trial data store that contains data from multiple heterogeneous clinical trials, all in the normalized attribute vector formal;

creating, in a graph database, a graph database instance for a set of the normalized attribute vectors stored in the data store, including:

adding, to the graph database instance, respective nodes corresponding to the normalized attribute vectors of the set; and

for each of the normalized attribute vectors:

comparing the model attributes with the model attributes of normalized attribute vectors corresponding to other nodes in the graph database instance using a first similarity score;

creating respective edges in the graph database instance between nodes corresponding to compared normalized attribute vectors having a first similarity score within a first predetermined threshold;

comparing the experiment attributes with the experiment attributes of normalized attribute vectors corresponding to the other nodes in the graph database instance using a second similarity score; and

creating respective edges in the graph database instance between nodes corresponding to compared normalized attribute vectors having a second similarity score within a second predetermined threshold;

creating, in a category database, a category database instance for the set of normalized attribute vectors, including, for each of the normalized attribute vectors:

comparing the model attributes to model attributes associated with categories of other normalized attribute vectors in the category database instance; and

associating, in the category database instance, the normalized attribute vector with categories corresponding to model attributes of the other normalized attribute vectors in the category database instance that match the model attributes of the normalized attribute vector;

creating a causal inference model from data in the clinical trial data store, the graph database instance, and the category database instance corresponding to the normalized attribute vectors; and

performing a causal analysis on the causal inference model.

14 . The system of claim 13 , the operations further comprising:

creating an empty normalized attribute form record, populating the created record with in-field patient data, backfilling the created record with at least some data from the clinical trial data for a particular drug and at least some kinetic model data of the particular drug, and loading the created and populated record into the clinical trial data store.

15 . The system of claim 14 , further comprising:

checking the consistency of the created record, including:

computing the response of a patient to the particular drug based on the in-field patient data;

comparing the computed response with a response predicted by the kinetic model data of the particular drug; and

if the computed response is within a predetermined threshold of the predicted response, indicating that the created record is verified and can be loaded into the clinical trial data store.

16 . A method to perform causal analysis on data relating to Anti-Microbial Resistant (AMR) bacteria, comprising:

receiving in-field patient data, clinical trial data, and bacterial kinetic model data, wherein the in-field patient data includes response curve data for a particular drug, and the bacterial kinetic model data includes data about biological markers not necessarily related to AMR and time limit threshold data for the biological markers;

converting the in-field patient data of AMR infection instances into normalized attribute vectors having a normalized attribute vector form;

backfilling the converted in-field patient data with the bacterial kinetic model data;

verifying the consistency of the backfilled in-field patient data by computing the response curve data and comparing the response curve data against the bacterial kinetic model data;

loading at least some of the backfilled in-patient data based on the verification into a clinical trial data store;

loading an experiment and the clinical trial data into the clinical trial data store;

loading a graph database and a category database with the normalized attribute vectors corresponding to the clinical trial data loaded into the clinical trial data store;

extracting from the clinical trial data store, graph database, and category database a data model based on the normalized attribute vectors for performing analytics;

performing causal analysis on the extracted data model.