Methods and systems for assessing inflammatory disease with deep learning
The present disclosure provides methods and systems of identifying an inflammatory disease or condition, e.g., an inflammatory bowel disease in a subject using a DeepLearning model. The DeepLearning model may be used to predict, treat, monitor, and/or prevent the inflammatory disease or condition in the subject, as well as to characterize a subtype of the inflammatory disease or condition.
1 . A method for determining a risk of developing Crohn's disease in a subject, comprising:
(a) obtaining a biological sample from the subject;
(b) selectively amplifying and tagging a plurality of target DNA molecules from the biological sample;
(c) sequencing or genotyping the plurality of amplified and tagged target DNA molecules from the biological sample from the subject to generate a dataset comprising genetic data;
(d) generating, based at least on the dataset comprising the genetic data, a Crohn's disease profile of the biological sample of the subject;
wherein the Crohn's disease profile is generated to include a difference between a plurality of feature sets from different timepoints;
wherein the generating the Crohn's disease profile includes processing, using one or more data processor, the dataset to determine, for inclusion in each feature set of the plurality of feature sets, one or more quantitative measures indicative of a presence, an absence, and/or a relative quantity of a plurality of known susceptibility variants at the timepoint of the biological sample; and
wherein the plurality of known susceptibility variants include rs5743293, rs2066845, and rs3116496;
(e) applying, using the one or more data processors, a deep learning prediction model trained to account for one or more non-linear, higher order combinatorial effects of the plurality of known susceptibility variants within a single timepoint and across the plurality of timepoints;
wherein the deep learning prediction model is applied to the Crohn's disease profile to determine a change in a risk of developing Crohn's disease in the subject;
wherein the deep learning prediction model includes a plurality of data pre-processing layers followed by one or more fully-connected layers; and
wherein the plurality of pre-processing layers down-sample each feature set to reduce a dimension of each feature set; and
(f) administering, based at least on the change in the risk of developing Crohn's disease in the subject, a therapeutic intervention including one or more immunomodulators, wherein a different immunomodulator than a current immunomodulator of the subject is administered in response to the change in risk of developing the Crohn's disease comprising an elevated risk.
2 . The method of claim 1 , wherein the deep learning prediction model comprises a deep learning algorithm comprises one or more of a neural network, a random forest model, and a gradient-boosted model.
3 . The method of claim 1 , wherein the Crohn's disease profile is further generated by at least applying the deep learning prediction model to a set of clinical health data of the subject, and wherein the clinical health data of the subject comprises having or not having a familial history of an inflammatory disease or disorder, having or not having hypertension or pre-hypertension, having or not having diabetes or pre-diabetes, being or not being overweight or obese, smoking or not smoking, drinking or not drinking alcohol, using or not using drugs, or any combination of one or more thereof.
4 . The method of claim 1 , wherein the Crohn's disease profile further comprises one or more serological markers.
5 . The method of claim 1 , wherein the change in the risk of developing Crohn's disease is determined at a sensitivity of at least about 70%.
6 . The method of claim 1 , wherein the change in the risk of developing Crohn's disease is determined at a specificity of at least about 70%.
7 . The method of claim 1 , wherein the change in the risk of developing Crohn's disease is determined at a positive predictive value of at least about 70%.
8 . The method of claim 1 , wherein the change in the risk of developing Crohn's disease is determined at a negative predictive value of at least about 70%.
9 . The method of claim 1 , wherein the change in the risk of developing Crohn's disease is determined with an Area Under Curve (AUC) of at least about 0.70.
10 . The method of claim 1 , further comprising:
training the deep learning prediction model using a first set of independent training samples associated with a presence of the Crohn's disease and a second set of independent training samples associated with an absence of the Crohn's disease.
11 . The method of claim 1 , further comprising optimizing a set of hyperparameters of the deep learning prediction model.
12 . The method of claim 11 , wherein the set of hyperparameters is optimized by at least performing a grid search.
13 . The method of claim 11 , wherein the optimizing the set of hyperparameters includes optimizing a number of layers and/or a number of neurons of a neural network comprising the deep learning prediction model.
14 . The method of claim 1 , wherein the deep learning prediction model comprises an ensemble of a plurality of neural networks, random forest models, and/or gradient boosted models.
15 . The method of claim 4 , wherein the one or more serological markers comprise ANCA, CBirl, IgA-ASCA, or IgG-ASCA, or a combination of two or more thereof.
16 . The method of claim 1 , wherein the therapeutic intervention further includes one or more of (i) terminating the current therapeutic intervention of the subject and (ii) a secondary clinical test where the change in risk of developing the Crohn's disease comprises an elevated risk.
17 . The method of claim 1 , wherein the therapeutic intervention comprises one or more of (i) continuing the current therapeutic intervention of the subject and (ii) a secondary clinical test where the change in the risk of developing Crohn's disease comprises a decreased risk of Crohn's disease.
18 . The method of claim 1 , further comprising:
applying, using the one or more data processors, the deep learning prediction model to the Crohn's disease profile to determine, based at least on the difference between the plurality of feature sets, an efficacy or a non-efficacy of a current therapeutic intervention of the subject; and administering, based at least on the non-efficacy of the current therapeutic intervention of the subject, the different therapeutic intervention instead of the current therapeutic intervention.
19 . The method of claim 1 , wherein the one or more immunomodulators include an anti-tumor necrosis factor alpha (TNFa) therapy, a glucocorticosteriod, an anti-TNF therapy, an anti-a4-b7 therapy, an anti-IL12p40 therapy, Thalidomide, and/or Cytoxin.