Precision treatment with machine learning and digital twin technology for optimal metabolic outcomes

1. A method for generating a recommendation for improving a metabolic state for a patient, the method comprising:

accessing a plurality of biosignals recorded for the patient, the plurality of biosignals comprising sensor data collected during a current time period by one or more wearable sensors worn by the patient;

encoding a subset of the biosignals collected during the current time period into a vector representation;

inputting the vector representation into a patient-specific metabolic model to determine a metabolic state of the patient at a conclusion of the current time period, wherein the patient-specific metabolic model is iteratively trained based on a training dataset of previous metabolic states of the patient and historical biosignals contributing to each previous metabolic state, the patient-specific metabolic model comprising:

a set of parameter values determined based on labels assigned to the previous metabolic states, each parameter value of the set describing a weight associated with the plurality of biosignals; and

a function for determining a risk score representing one or more effects of the plurality of biosignals on a personalized metabolic profile of the patient;

and

generating a patient-specific treatment recommendation outlining objectives for the patient to complete to improve the determined metabolic state of the patient.

2. The method of claim 1 , further comprising:

accessing a population-level training dataset with labeled metabolic states determined for a population of patients, and with recorded sensor data points and lab test data points that contributed to each labeled metabolic state;

training a baseline metabolic model based on the population-level training dataset, wherein the trained baseline metabolic model determines a metabolic state of an average patient of the population of patients; and

generating the patient-specific metabolic model by training the baseline metabolic model based on the training dataset of previous metabolic states of the patient and historical biosignals contributing to each previous metabolic state, wherein the patient-specific metabolic model is a version of the baseline metabolic model personalized for the patient.

3. The method of claim 1 , further comprising:

updating the training dataset with the determined metabolic state of the patient, a label assigned to the determined metabolic state, and the plurality of biosignals recorded for the patient across the current time period; and

re-training the patient-specific metabolic model based on the updated training dataset.

4. The method of claim 1 , further comprising:

receiving, by user input to an application on a patient device, a recorded timeline of patient data for the current time period, patient data comprising one or more of: 1) foods consumed by the patient throughout the current time period, 2) medications taken by the patient throughout the current time period, and 3) symptoms experienced by the patient throughout the current time period;

encoding the recorded timeline of patient data into a vector representation; and

generating a prediction of the metabolic state of the patient across multiple time points of the current time period by inputting the personalized metabolic profile of the patient and the vector representation of the recorded timeline of patient data into the patient-specific metabolic model.

5. The method of claim 4 , wherein the patient-specific metabolic model is trained to generate the prediction of the metabolic state of the patient based on a second training dataset of previous metabolic states of the patient and a history of recorded patient data contributing to each previous metabolic state.

6. The method of claim 4 , wherein the patient-specific metabolic model generates a prediction of the metabolic state of the patient based a relative glucose impact of each food consumed by the patient during the current time period and classifies each food consumed by the patient based on the relative glucose impact the food.

7. The method of claim 4 , wherein the patient-specific metabolic model generates a prediction of the metabolic state of the patient based on an average blood glucose level of the patient over a 24-hour calendar day given the personalized metabolic profile of the patient and the foods consumed by the patient throughout the current time period.

8. The method of claim 1 , wherein generating the patient-specific treatment recommendation comprises:

inputting the plurality of biosignals and the determined metabolic state of the patient to a rule-based model, wherein the rule-based model codifies a set of medical information into a set of rules defined by medical experts; and

generating, by the rule-based model, a set of instructions to improve the determined metabolic state of the patient that are consistent with the rules defined by medical experts.

9. The method of claim 1 , wherein generating the patient-specific treatment recommendation comprises:

inputting the plurality of biosignals and the determined metabolic state of the patient to a cohorting model to identify a group of patients with similar metabolic states, the group identified based on a comparison of personalized metabolic profiles for each patient of the group to the determined metabolic state of the patient;

determining a representative metabolic profile for the group of patients, the representative metabolic profile based on an aggregate of personalized metabolic profiles for each patient in the group and the determined metabolic state of the patient; and

generating the patient-specific treatment recommendation for the group of patients based on the representative metabolic profile for the group of patients.

10. The method of claim 1 , further comprising:

receiving, by user input to an application on a patient device, a recorded timeline of patient data for the current time period, patient data comprising one or more of: 1) foods consumed by the patient throughout the current time period, 2) medications taken by the patient throughout the current time period, and 3) symptoms experienced by the patient throughout the current time period; and

comparing the recorded timeline of patient data with the determined metabolic state of the patient to determine an effect of each food item, medication, and symptom on the metabolic profile of the patient.

11. A non-transitory computer readable medium storing instructions for generating a recommendation for improving a metabolic state for a patient encoded thereon that, when executed by a processor, cause the processor to:

access a plurality of biosignals recorded for the patient, the plurality of biosignals comprising sensor data collected during a current time period by one or more wearable sensors worn by the patient;

encode a subset of the biosignals collected during the current time period into a vector representation;

input the vector representation into a patient-specific metabolic model to determine a metabolic state of the patient at a conclusion of the current time period, wherein the patient-specific metabolic model is iteratively trained based on a training dataset of previous metabolic states of the patient and historical biosignals contributing to each previous metabolic state, the patient-specific metabolic model comprising:

a set of parameter values determined based on labels assigned to the previous metabolic states, each parameter value of the set describing a weight associated with the plurality of biosignals; and

a function for determining a risk score representing one or more effects of the plurality of biosignals on a personalized metabolic profile of the patient;

and

generate a patient-specific treatment recommendation outlining objectives for the patient to complete to improve the determined metabolic state of the patient.

12. The non-transitory computer readable medium of claim 11 , wherein the instructions further cause the processor to:

receive, by user input to an application on a patient device, a recorded timeline of patient data for the current time period, patient data comprising one or more of: 1) foods consumed by the patient throughout the current time period, 2) medications taken by the patient throughout the current time period, and 3) symptoms experienced by the patient throughout the current time period;

encode the recorded timeline of patient data into a vector representation; and

generate a prediction of the metabolic state of the patient across multiple time points of the current time period by inputting the personalized metabolic profile of the patient and the vector representation of the recorded timeline of patient data into the patient-specific metabolic model.

13. The non-transitory computer readable medium of claim 11 , wherein the instructions for generating the patient-specific treatment recommendation further cause the processor to:

input the plurality of biosignals and the determined metabolic state of the patient to a rule-based model, wherein the rule-based model codifies a set of medical information into a set of rules defined by medical experts; and

generate, by the rule-based model, a set of instructions to improve the determined metabolic state of the patient that are consistent with the rules defined by medical experts.

14. The non-transitory computer readable medium of claim 11 , wherein the instructions for generating the patient-specific treatment recommendation further cause the processor to:

input the plurality of biosignals and the determined metabolic state of the patient to a cohorting model to identify a group of patients with similar metabolic states, the group identified based on a comparison of personalized metabolic profiles for each patient of the group to the determined metabolic state of the patient;

determine a representative metabolic profile for the group of patients, the representative metabolic profile based on an aggregate of personalized metabolic profiles for each patient in the group and the determined metabolic state of the patient; and

generate the patient-specific treatment recommendation for the group of patients based on the representative metabolic profile for the group of patients.

15. The non-transitory computer readable medium of claim 11 , wherein the instructions further cause the processor to:

receive, by user input to an application on a patient device, a recorded timeline of patient data for the current time period, patient data comprising one or more of: 1) foods consumed by the patient throughout the current time period, 2) medications taken by the patient throughout the current time period, and 3) symptoms experienced by the patient throughout the current time period; and

compare the recorded timeline of patient data with the determined metabolic state of the patient to determine an effect of each food item, medication, and symptom on the metabolic profile of the patient.

16. A system comprising:

one or more wearable sensors worn by a patient, each of the one or more wearable sensors configured to collected sensor data during a current time period;

an application stored on a patient device that presents metabolic insights generated for the patient; and

a non-transitory computer readable medium storing instructions for generating a recommendation for improving a metabolic state for a patient encoded thereon that, when executed by a processor, cause the processor to:

access a plurality of biosignals collected during the current time period by one or more wearable sensors worn by the patient;

encode a subset of the biosignals into a vector representation;

input the vector representation into a patient-specific metabolic model to determine a metabolic state of the patient at a conclusion of the current time period, wherein the patient-specific metabolic model is iteratively trained based on a training dataset of previous metabolic states of the patient and historical biosignals contributing to each previous metabolic state, the patient-specific metabolic model comprising:

a set of parameter values determined based on labels assigned to the previous metabolic states, each parameter value of the set describing a weight associated with the plurality of biosignals; and

a function for determining a risk score representing one or more effects of the plurality of biosignals on a personalized metabolic profile of the patient;

and

generate a patient-specific treatment recommendation outlining objectives for the patient to complete to improve the determined metabolic state of the patient.

17. The system of claim 16 , wherein the instructions further cause the processor to:

receive, by user input to an application on a patient device, a recorded timeline of patient data for the current time period, patient data comprising one or more of: 1) foods consumed by the patient throughout the current time period, 2) medications taken by the patient throughout the current time period, and 3) symptoms experienced by the patient throughout the current time period;

encode the recorded timeline of patient data into a vector representation; and

generate a prediction of the metabolic state of the patient across multiple time points of the current time period by inputting the personalized metabolic profile of the patient and the vector representation of the recorded timeline of patient data into the patient-specific metabolic model.

18. The system of claim 16 , wherein the instructions for generating the patient-specific treatment recommendation further cause the processor to:

input the plurality of biosignals and the determined metabolic state of the patient to a rule-based model, wherein the rule-based model codifies a set of medical information into a set of rules defined by medical experts; and

generate, by the rule-based model, a set of instructions to improve the determined metabolic state of the patient that are consistent with the rules defined by medical experts.

19. The system of claim 16 , wherein the instructions for generating the patient-specific treatment recommendation further cause the processor to:

input the plurality of biosignals and the determined metabolic state of the patient to a cohorting model to identify a group of patients with similar metabolic states, the group identified based on a comparison of personalized metabolic profiles for each patient of the group to the determined metabolic state of the patient;

determine a representative metabolic profile for the group of patients, the representative metabolic profile based on an aggregate of personalized metabolic profiles for each patient in the group and the determined metabolic state of the patient; and

generate the patient-specific treatment recommendation for the group of patients based on the representative metabolic profile for the group of patients.

20. The system of claim 16 , wherein the instructions further cause the processor to:

receive, by user input to an application on a patient device, a recorded timeline of patient data for the current time period, patient data comprising one or more of: 1) foods consumed by the patient throughout the current time period, 2) medications taken by the patient throughout the current time period, and 3) symptoms experienced by the patient throughout the current time period; and

compare the recorded timeline of patient data with the determined metabolic state of the patient to determine an effect of each food item, medication, and symptom on the metabolic profile of the patient.