Computer systems and methods for inferring scores for health metrics
Provided herein are computer systems and method for producing models that infer health scores for health metrics for a subject. A first learning algorithms is trained using raw feature data stored in computer memory to build a first model that infers feature cluster scores for each of a plurality of feature clusters in a feature group. A second learning algorithms is trained using the inferred feature cluster scores stored in computer memory to build a second model that infers a heath score for a health metric.
1 . A method that infers a health score for a health metric for a subject comprising:
(a) providing a first data set comprising, for the subject, feature data for each of a plurality of features selected from one or more feature groups;
(b) defining a plurality of feature clusters from the feature data, wherein each feature cluster in the plurality of feature clusters comprises a collection of features in the plurality of features;
(c) executing a first computer model on the first data set to assign feature cluster scores for the health metric to each of the plurality of feature clusters, wherein the first computer model comprises a first machine learning algorithm, wherein each feature cluster score is a quantitative measure, thereby generating a second data set comprising a plurality of feature cluster scores;
(d) executing a second computer model on the second data set comprising the plurality of feature cluster scores to infer a health score for the health metric for the subject, wherein the second computer model comprises a second machine learning algorithm;
(e) obtaining, from the second computer model, a health score for the health metric for the subject that indicates a severity of the health metric that is present in the subject, wherein the severity is higher than a lowest category of severity; and
(f) administering, responsive to determining the health score for the subject indicating the severity of the health metric is higher than the lowest category of severity, an intervention to the subject to improve the health score, thereby reducing the severity of the health metric in the subject, wherein the health metric is inflammation, and wherein the intervention comprises increased consumption of one or more of probiotics, fibers, and polyphenols.
2 . The method of claim 1 , further comprising training the first computer model and the second computer model by a process comprising:
a) receiving into computer memory a first training set, wherein the first training data set comprises, for each of a plurality of subjects, (1) feature data for each of a plurality of features selected from one or more feature groups and (2) feature cluster labels for each of one or a plurality of feature clusters,
b) with a computer processor, training the first machine learning algorithm on the first training data set, wherein the first machine learning algorithm develops the first model that infers cluster scores for each of a plurality of feature clusters;
c) with a computer processor, executing the first model on a test data set comprising, for each of a plurality of subjects, feature data for the features, to produce a cluster score data set comprising, for each of the plurality of subjects in the test data set, feature cluster scores for each of the plurality of feature clusters;
d) labeling each subject in the cluster score data set with a health label for the health metric to produce a second training data set; and
e) with a computer processor, training the second machine learning algorithm on the second training data set to develop the second model that infers a health score for the health metric based on feature cluster scores.
3 . The method of claim 2 , wherein the feature cluster labels comprise partial order cluster rankings assigned by a first person skilled in the field.
4 . The method of claim 3 , wherein the health labels comprise partial order health rankings assigned by a second person skilled in the field.
5 . The method of claim 1 , wherein the subject is a human subject.
6 . The method of claim 1 , wherein the feature data comprises microbiome feature data and phenotype feature data.
7 . The method of claim 1 , wherein the feature groups comprise gene expression data, microbial taxa data and phenotypic data and the feature data includes at least:
(1) data on gene expression for each of a plurality of genes in a microbiome of the subject;
(2) microbiome taxa quantity data for a plurality of microbes in a microbiome of the subject; and
(3) phenotypic data for a plurality of different phenotypic traits of the subject.
8 . The method of claim 7 , wherein the microbiome is a fecal microbiome.
9 . The method of claim 7 , wherein the gene expression data comprises metatranscriptome sequence information.
10 . The method of claim 7 , wherein the gene expression data comprises data on expression of at least any of 10, 50, 100, 150, 200, 500, or 1000 different genes.
11 . The method of claim 7 , wherein the gene expression data comprises data on expression of genes involved in pathways associated with the health metric.
12 . The method of claim 7 , wherein the microbiome taxa data comprises data on microbes belonging to at least any of 10, 50, 100, 150, 200, 500, or 1000 different taxa.
13 . The method of claim 7 , wherein the microbiome taxa data comprises data on one or more groups selected from bacteria, viruses, Archaebacteria, yeast, fungi, parasites and bacteria phages.
14 . The method of claim 7 , wherein the phenotypic data for the plurality of different phenotypic traits comprises subject responses to one or more phenotypic traits selected from the group consisting of: age, sex, weight, blood type, headaches, faintness, dizziness, insomnia, watery or itchy eyes, swollen, red or sticky eyelids, bags or dark circles under eyes, blurred or tunnel vision, not including near or far-sightedness, itchy ears, earaches, ear infections, drainage from ear, ringing in ears, hearing loss, stuffy nose, sinus problems, hay fever, sneezing attacks, excessive mucus formation, chronic coughing, gagging, need to clear throat, sore throat, hoarseness, loss of voice, swollen or discolored tongue, gums or lips, canker sores, acne, hives, rashes, dry skin, hair loss, flushing, hot flashes, excessive sweating, irregular or skipped heartbeat, rapid or pounding heartbeat, chest pain, chest congestion, asthma, bronchitis, shortness of breath, difficulty breathing, bloated feeling, nausea, vomiting, diarrhea, constipation, belching, passing gas, heartburn, intestinal pain, stomach pain, pain or aches in joints, arthritis, stiffness or limitation of movement, pain or aches in muscles, feeling of weakness or tiredness, binge eating, binge drinking, craving certain foods, excessive weight, compulsive eating, water retention, underweight, fatigue, sluggishness, apathy, lethargy, hyperactivity, restlessness, poor memory, confusion, poor comprehension, poor concentration, poor physical coordination, difficulty in making decisions, stuttering or stammering, slurred speech, learning disabilities, poor physical coordination or clumsiness, numbness or tingling in hands or feet, mood swings, anxiety, fear or nervousness, anger, irritability or aggressiveness, sadness or depression, frequent illness such as colds, frequent or urgent urination, genital itch or discharge, decreased libido and PMS.
15 . The method of claim 1 , wherein the feature clusters comprise a plurality of gene clusters, a plurality of microbial taxa clusters and a plurality of phenotype clusters.
16 . The method of claim 1 , wherein the feature cluster score is a quantity having a discrete or continuous range.
17 . The method of claim 1 , wherein the feature data is provided by:
(i) providing a biological sample from the subject comprising microbiota;
(ii) sequencing nucleic acids in the biological sample to produce sequence data; and
(iii) determining data for gene expression and microbiome taxa quantities using the sequence data.
18 . The method of claim 1 , wherein the second computer model generates a positive health component and a negative health component and combines the components to produce the health score for the health metric.
19 . The method of claim 1 , wherein feature clusters comprise one or more of: pro-inflammatory gene expression, pro-inflammatory taxa amounts, anti-inflammatory gene expression, anti-inflammatory taxa amounts, and intestinal barrier insufficiency gene expression and intestinal barrier insufficiency taxa amounts.
20 . The method of claim 1 , further comprising obtaining a second health score for a second health metric for the subject, wherein the second health metric is metabolic fitness, and feature clusters comprise one or more of:
(i) gene expression in pathways selected from one or more of: secondary bile acid pathway, primary bile acid pathway, butyrate pathway, methanogenesis pathway, acetate pathway, propionate pathway, branch chain amino acid pathway, long chain fatty acid metabolism pathway and long chain carbohydrate metabolic pathway; and
(ii) taxa clusters selected from one or more of: Prevotella (genus)/ Bacteroides (genus) ratio, Eubacterium rectale (species), Eubacterium eligens (species), Faecalibacterium prausnitzii (species), Akkermansia muciniphila (species), metabolic-related probiotic species (functional group), Roseburia (genus), Bifidobacterium (genus), Lactobacillus (genus), Clostridium butyricum (species), Allobaculum (genus), Firmicutes (phylum)/Bacteroidetes (phylum) ratio, Lachnospiraceae (family), Enterobacteriaceae (family), Ralstonia pickettii (species), Bilophila wadsworthia (species).
21 . The method of claim 1 , wherein the first machine learning algorithm or the second machine learning algorithm use supervised methods selected from the group consisting of artificial neural networks, decision trees, random forests, discriminant analyses, linear classifiers, partial least squares (PLS) regression, principal components regression (PCR), mixed or random-effects models, non-parametric classifiers, support vector machines, and ensemble methods.
22 . The method of claim 1 , wherein the health score obtained from the second computer model indicates that the severity of the health metric that is present in the subject is actionable.
23 . The method of claim 1 , wherein the health score obtained from the second computer model is a discrete number on a scale from 1 to 10, and wherein the health score for the subject is from 4 to 6 or from 7 to 10.
24 . A method comprising:
a) training a first machine learning algorithm on a first training data set,
wherein the first training data set comprises, for each of a plurality of objects, (1) feature data for each of a plurality of features and (2) a feature cluster label for each of one or a plurality of feature clusters, wherein the plurality of feature clusters is defined from the feature data and each feature cluster in the plurality of feature clusters comprises a collection of features in the plurality of features, and
wherein the first machine learning algorithm develops a first model that infers a cluster score for each of the feature clusters based on the feature data, wherein the cluster score is a quantitative measure;
b) executing the first model on a test data set comprising, for each of a plurality of objects, feature data for the features, to produce a cluster score data set comprising, for each of the plurality of objects in the test data set, a feature cluster scores for each of the feature clusters;
c) labeling each subject in the cluster score data set with a label for a categorical variable to produce a second training data set;
d) training a second machine learning algorithm on the second training data set to develop a second model that infers a label for the categorical variable based on feature cluster scores;
e) using the first model and the second model to obtain a label for the categorical variable for a test subject, wherein the label for the categorical variable is an indication of a severity of inflammation present in the test subject, wherein the severity is higher than a lowest category of severity; and
f) administering, responsive to the label for the categorical variable for the test subject indicating the severity of inflammation is higher than the lowest category of severity, an intervention to the test subject to improve the label for the categorical variable, thereby reducing the severity of inflammation, wherein the intervention comprises increased consumption of one or more of probiotics, fibers, and polyphenols.