Systems and methods for topology-based clinical data mining

1. A system for topology-based clinical data mining, the system comprising:

at least one processor; and

a memory storing processor-executable codes, wherein upon executing the processor-executable codes the at least one processor is configured to:

process clinical datasets and generate:

a first table, the first table comprising first rows representing trial subjects and first columns including outcomes of the trial subjects; and

a second table, the second table comprising rows representing the trial subjects and columns including predictors of the trial subjects;

generate, based on the first table, a plurality of metric graphs, wherein:

each metric graph of the plurality of metric graphs includes a same set of nodes, each of the nodes corresponding to a single trial subject of the trial subjects in the first table; and

generating a set of edges in a metric graph of the plurality of metric graphs includes:

selecting a criterion from a plurality of criteria for selectively connecting the nodes; and

selectively connecting the nodes by edges according to the selected criterion;

select, from the plurality of metric graphs and based on a further criterion, a graph of interest;

generate a compressed version of the graph of interest, the compressed version including a clustered graph;

generate a first layout of the graph of interest and a second layout of the compressed version of the graph of interest, the first layout and the second layout being visually aligned;

display, via a graphical user interface, based on one of the first layout or the second layout, a graphical representation of the graph of interest;

perform, using one or more machine learning algorithms, an automatic search to identify at least one group of related trial subjects;

highlight nodes in the graphical representation, the nodes corresponding to the related trial subjects;

receive, via the graphical representation, a user input, the user input including one or more selected groups of the trial subjects;

perform, using the second table, a statistical analysis of predictors associated with trial subjects within the one or more selected groups of the trial subjects; and

display a report with results of the statistical analysis.

2. The system of claim 1 , wherein the outcomes include two or more of the following: a biomarker, a vital sign, a result of physiological measurement, and a questionnaire item.

3. The system of claim 1 , wherein the predictors include one or more of the following: a demographic attribute, a medical history attribute, and a medical interventions attribute.

4. The system of claim 1 , wherein the at least one processor is configured to normalize data in the clinical datasets.

5. The system of claim 1 , wherein the at least one processor is configured to fill in missing values for the outcomes in the first table.

6. The system of claim 1 , wherein the at least one processor is configured to generate, based on the clinical datasets, one or more synthetic variables, the one or more synthetic variables being a combination of one or more outcomes associated with the trial subjects.

7. The system of claim 1 , wherein the criterion includes a determination that a distance between data points corresponding to the trial subjects does not exceed a pre-determined distance threshold, the data points including vectors of the outcomes of the trial subjects.

8. The system of claim 7 , wherein the distance is determined by a distance function including one of a Euclidean distance, a normalized Euclidean distance, a Manhattan distance, a Hamming distance, and a Gower distance.

9. The system of claim 1 , wherein the at least one processor is configured to:

determine a graph with a highest value of an objective function of the plurality of metric graphs; and

select the graph with the highest value of the objective function as the graph of interest.

10. The system of claim 1 , wherein the statistical analysis includes calculating p-values for statistical tests to determine whether a distribution of values of one of the predictors of the trial subjects within a first group from the one or more selected groups is different from a distribution of values of the one of the predictors for the trial subjects within a second group from the one or more selected groups.

11. The system of claim 1 , wherein the statistical analysis includes calculating p-values for statistical tests to determine whether a distribution of values of one of the predictors for the trial subjects within the one or more selected groups is different from a distribution of values of the one of the predictors for the rest of the trial subjects.

12. The system of claim 1 , wherein the at least one processor is configured to:

receive a further user input, the further user input including a selected predictor from the predictors or a selected outcome from the outcomes; and

assign colors to nodes of the graph of interest, the colors being determined based on projection values or values for the selected predictor or the selected outcome, wherein the projection values are determined based on data points including vectors of the outcomes of trial subjects corresponding to the nodes.

13. A method for topology-based clinical data mining, the method comprising:

processing, by a pre-processing module, clinical datasets to generate a first table, the first table comprising first rows representing trial subjects and first columns including outcomes of the trial subjects, and a second table, the second table comprising second rows representing the trial subjects and second columns including predictors of the trial subjects;

generating, by a graph construction module and based on the first table, a plurality of metric graphs, wherein:

each metric graph of the plurality of metric graphs includes a same set of nodes, each of the nodes corresponding to a single trial subject of the trial subjects in the first table; and

generating a set of edges in a metric graph of the plurality of metric graphs includes:

selecting a criterion from a plurality of criteria for selectively connecting the nodes; and

selectively connecting the nodes by edges according to the selected criterion;

selecting, by the graph construction module, from the plurality of metric graphs and based on a further criterion, a graph of interest;

generating, by the graph construction module, a compressed version of the graph of interest, the compressed version including a clustered graph;

generating, by the graph construction module, a first layout of the graph of interest and a second layout of the compressed version of the graph of interest, the first layout and the second layout being visually aligned;

displaying, by an interactive visualization module and based on the first layout or the second layout, a graphical representation of the graph of interest;

assigning, by the interactive visualization module, colors to the nodes of the graph of interest, wherein the colors are determined based on projection values or values of a selected outcome or a selected predictor, wherein the projections values are determined based on data points including vectors of the outcomes of trial subjects corresponding to the nodes;

performing, by the interactive visualization module and using one or more machine learning algorithms, an automatic search to identify at least one group of related trial subjects;

highlighting, by the interactive visualization module, nodes in the graphical representation, the nodes corresponding to the related trial subjects;

receiving, by the interactive visualization module and via the graphical representation, a user input including one or more selected groups of the trial subjects;

performing, by the interactive visualization module and using the second table, a statistical analysis of predictors associated with trial subjects within the one or more selected groups of the trial subjects; and

displaying a report with results of the statistical analysis.

14. The method of claim 13 , wherein:

the outcomes include two or more of the following: a biomarker, a vital sign, a result of a physiological measurement, and a questionnaire item; and

the predictors include one or more of the following: a demographic attribute, a medical history attribute, and a medical interventions attribute.

15. The method of claim 13 , wherein processing the clinical datasets includes one or more of the following:

normalizing data in the clinical datasets;

filling in missing values for the outcomes in the first table; and

generating, based on the clinical datasets, synthetic variables, the synthetic variables representing combinations of one or more outcomes associated with the trial subjects.

16. The method of claim 13 , wherein the criterion includes a determination that a distance between the data points corresponding to the trial subjects does not exceed a pre-determined distance threshold, the data points including vectors of the outcomes of the trial subjects.

17. The method of claim 13 , wherein selecting the graph of interest includes:

determining a graph with a highest value of an objective function of the plurality of metric graphs; and

selecting the graph with the highest value of the objective function as the graph of interest.

18. The method of claim 13 , wherein the statistical analysis includes calculating p-values for statistical tests to determine whether a distribution of values of one of the predictors for the trial subjects within a first group from the one or more selected groups is different from a distribution of the values of the one of the predictors for the trial subjects within a second group from the one or more selected groups.

19. The method of claim 13 , wherein the statistical analysis includes calculating p-values for statistical tests to determine whether a distribution of values of one of the predictors for the trial subjects within a group from the one or more selected groups is different from a distribution of values of the one of the predictors for the rest of the trial subjects within the clinical datasets.

20. A system for topology-based clinical data mining, the system comprising:

at least one processor; and

a memory storing processor-executable codes, wherein upon executing the processor-executable codes the at least one processor is configured to:

process clinical datasets to generate a first table, the first table comprising first rows representing trial subjects and first columns including outcomes of the trial subjects, and a second table, the second table comprising second rows representing the trial subjects and second columns including predictors of the trial subjects;

generate, based on the first table, a plurality of metric graphs, wherein:

each metric graph of the plurality of metric graphs includes a same set of nodes and edges, each of the nodes corresponding to a single trial subject of the trial subjects in the first table; and

generating a set of edges in a metric graph of the plurality of metric graphs includes:

selecting a criterion from a plurality of criteria for selectively connecting the nodes; and

selectively connecting the nodes by edges according to the selected criterion;

estimate values of an objective function of the plurality of metric graphs;

select, from the plurality of metric graphs and based on the values of the objective function, a graph of interest;

generate a compressed version of the graph of interest, the compressed version including a clustered graph;

generate a first layout of the graph of interest and a second layout of the compressed version of the graph of interest, the first layout and the second layout being visually aligned;

display, via a graphical user interface, based on one of the first layout and the second layout, a graphical representation of the graph of interest;

assign, based on projection values or values for a selected predictor or a selected outcome, colors to the nodes of the graph of interest, wherein the projection values are determined based on data points including vectors of the outcomes of the trial subjects corresponding to the nodes;

perform, using one or more machine learning algorithms, an automatic search to identify at least one group of related trial subjects;

highlight nodes in the graphical representation, the nodes corresponding to the related trial subjects;

receive, via the graphical representation, a user input, the user input including one or more selected groups of the trial subjects; and

perform, using the second table, a statistical analysis of predictors associated with the trial subjects, wherein the statistical analysis includes one of:

calculating p-values for statistical tests to determine whether a distribution of values of one of the predictors for the trial subjects within a first group from the one or more selected groups is different from a distribution of values of the one of the predictors for the trial subjects within a second group from the one or more selected groups;

calculating p-values for the statistical tests to determine whether a distribution of values of the one of the predictors for the trial subjects within a group from the one or more selected groups is different from a distribution of values of the one of the predictors for the rest of the trial subjects within the clinical datasets; and

displaying results of the statistical analysis.