Deep learning generates custom-made logistic regression models for explaining how breast cancer subtypes are classified.

Differentiating the intrinsic subtypes of breast cancer is crucial for deciding the best treatment strategy. Deep learning can predict the subtypes from genetic information more accurately than conventional statistical methods, but to date, deep learning has not been directly utilized to examine which genes are associated with which subtypes. To clarify the mechanisms embedded in the intrinsic subtypes, we developed an explainable deep learning model called a point-wise linear (PWL) model that generates a custom-made logistic regression for each patient. Logistic regression, which is familiar to both physicians and medical informatics researchers, allows us to analyze the importance of the feature variables, and the PWL model harnesses these practical abilities of logistic regression. In this study, we show that analyzing breast cancer subtypes is clinically beneficial for patients and one of the best ways to validate the capability of the PWL model. First, we trained the PWL model with RNA-seq data to predict PAM50 intrinsic subtypes and applied it to the 41/50 genes of PAM50 through the subtype prediction task. Second, we developed a deep enrichment analysis method to reveal the relationships between the PAM50 subtypes and the copy numbers of breast cancer. Our findings showed that the PWL model utilized genes relevant to the cell cycle-related pathways. These preliminary successes in breast cancer subtype analysis demonstrate the potential of our analysis strategy to clarify the mechanisms underlying breast cancer and improve overall clinical outcomes.

Potential progression biomarkers of diabetic kidney disease determined using comprehensive machine learning analysis of non-targeted metabolomics.

Diabetic kidney disease is the main cause of end-stage renal disease worldwide. The prediction of the clinical course of patients with diabetic kidney disease remains difficult, despite the identification of potential biomarkers; therefore, novel biomarkers are needed to predict the progression of the disease. We conducted non-targeted metabolomics using plasma and urine of patients with diabetic kidney disease whose estimated glomerular filtration rate was between 30 and 60 mL/min/1.73 m2. We analyzed how the estimated glomerular filtration rate changed over time (up to 30 months) to detect rapid decliners of kidney function. Conventional logistic analysis suggested that only one metabolite, urinary 1-methylpyridin-1-ium (NMP), was a promising biomarker. We then applied a deep learning method to identify potential biomarkers and physiological parameters to predict the progression of diabetic kidney disease in an explainable manner. We narrowed down 3388 variables to 50 using the deep learning method and conducted two regression models, piecewise linear and handcrafted linear regression, both of which examined the utility of biomarker combinations. Our analysis, based on the deep learning method, identified systolic blood pressure and urinary albumin-to-creatinine ratio, six identified metabolites, and three unidentified metabolites including urinary NMP, as potential biomarkers. This research suggests that the machine learning method can detect potential biomarkers that could otherwise escape identification using the conventional statistical method.

The PD-1 expression balance between effector and regulatory T cells predicts the clinical efficacy of PD-1 blockade therapies.

Immune checkpoint blockade has provided a paradigm shift in cancer therapy, but the success of this approach is very variable; therefore, biomarkers predictive of clinical efficacy are urgently required. Here, we show that the frequency of PD-1+CD8+ T cells relative to that of PD-1+ regulatory T (Treg) cells in the tumor microenvironment can predict the clinical efficacy of programmed cell death protein 1 (PD-1) blockade therapies and is superior to other predictors, including PD ligand 1 (PD-L1) expression or tumor mutational burden. PD-1 expression by CD8+ T cells and Treg cells negatively impacts effector and immunosuppressive functions, respectively. PD-1 blockade induces both recovery of dysfunctional PD-1+CD8+ T cells and enhanced PD-1+ Treg cell-mediated immunosuppression. A profound reactivation of effector PD-1+CD8+ T cells rather than PD-1+ Treg cells by PD-1 blockade is necessary for tumor regression. These findings provide a promising predictive biomarker for PD-1 blockade therapies.

A Method to Extract Feature Variables Contributed in Nonlinear Machine Learning Prediction.

BACKGROUND: Although advances in prediction accuracy have been made with new machine learning methods, such as support vector machines and deep neural networks, these methods make nonlinear machine learning models and thus lack the ability to explain the basis of their predictions. Improving their explanatory capabilities would increase the reliability of their predictions. OBJECTIVE: Our objective was to develop a factor analysis technique that enables the presentation of the feature variables used in making predictions, even in nonlinear machine learning models. METHODS: A factor analysis technique was consisted of two techniques: backward analysis technique and factor extraction technique. We developed a factor extraction technique extracted feature variables that was obtained from the posterior probability distribution of a machine learning model which was calculated by backward analysis technique. RESULTS: In evaluation, using gene expression data from prostate tumor patients and healthy subjects, the prediction accuracy of a model of deep neural networks was approximately 5% better than that of a model of support vector machines. Then the rate of concordance between the feature variables extracted in an earlier report using Jensen-Shannon divergence and the ones extracted in this report using backward elimination using Hilbert-Schmidt independence criteria was 40% for the top five variables, 40% for the top 10, and 49% for the top 100. CONCLUSION: The results showed that models can be evaluated from different viewpoints by using different factor extraction techniques. In the future, we hope to use this technique to verify the characteristics of features extracted by factor extraction technique, and to perform clinical studies using the genes, we extracted in this experiment.

A machine learning model to predict the risk of 30-day readmissions in patients with heart failure: a retrospective analysis of electronic medical records data.

BACKGROUND: Heart failure is one of the leading causes of hospitalization in the United States. Advances in big data solutions allow for storage, management, and mining of large volumes of structured and semi-structured data, such as complex healthcare data. Applying these advances to complex healthcare data has led to the development of risk prediction models to help identify patients who would benefit most from disease management programs in an effort to reduce readmissions and healthcare cost, but the results of these efforts have been varied. The primary aim of this study was to develop a 30-day readmission risk prediction model for heart failure patients discharged from a hospital admission. METHODS: We used longitudinal electronic medical record data of heart failure patients admitted within a large healthcare system. Feature vectors included structured demographic, utilization, and clinical data, as well as selected extracts of un-structured data from clinician-authored notes. The risk prediction model was developed using deep unified networks (DUNs), a new mesh-like network structure of deep learning designed to avoid over-fitting. The model was validated with 10-fold cross-validation and results compared to models based on logistic regression, gradient boosting, and maxout networks. Overall model performance was assessed using concordance statistic. We also selected a discrimination threshold based on maximum projected cost saving to the Partners Healthcare system. RESULTS: Data from 11,510 patients with 27,334 admissions and 6369 30-day readmissions were used to train the model. After data processing, the final model included 3512 variables. The DUNs model had the best performance after 10-fold cross-validation. AUCs for prediction models were 0.664 ± 0.015, 0.650 ± 0.011, 0.695 ± 0.016 and 0.705 ± 0.015 for logistic regression, gradient boosting, maxout networks, and DUNs respectively. The DUNs model had an accuracy of 76.4% at the classification threshold that corresponded with maximum cost saving to the hospital. CONCLUSIONS: Deep learning techniques performed better than other traditional techniques in developing this EMR-based prediction model for 30-day readmissions in heart failure patients. Such models can be used to identify heart failure patients with impending hospitalization, enabling care teams to target interventions at their most high-risk patients and improving overall clinical outcomes.

Machine-Learning Approach for Modeling Myelosuppression Attributed to Nimustine Hydrochloride.

PURPOSE: A major adverse effect arising from nimustine hydrochloride (ACNU) therapy for brain tumors is myelosuppression. Because its timing and severity vary among individual patients, the ACNU dose level has been adjusted in an empiric manner at individual medical facilities. To our knowledge, ours is the first study to develop a machine-learning approach to estimate myelosuppression through analysis of patient factors before treatment and attempts to clarify the relationship between myelosuppression and hematopoietic stem cells from daily clinical data. Adverse effect prediction will allow ACNU dose adjustment for patients predicted to have decreases in blood cell counts and will enable focused follow-up of patients undergoing chemoradiotherapy. PATIENTS AND METHODS: Patients were newly pathologically diagnosed with WHO grade 2 or 3 tumors and were treated with ACNU-based chemoradiotherapy. For detailed analysis of the timing and intensity of adverse effects in patients, we developed a data-weighted support vector machine (SVM) based on adverse event criteria (nadir-weighted SVM [NwSVM]). To evaluate the estimation accuracy of blood cell count dynamics, the determination coefficient ( r2) between real and estimated data was calculated by three regression methods: polynomial, SVM, and NwSVM. RESULTS: Only the NwSVM-based regression enabled estimation of the dynamics of all blood cell types with high accuracy (mean r2 = 0.81). The mean timing of nadir arrival estimated using this regression was 35 days for platelets, 41 days for RBCs, 52 days for lymphocytes, 57 days for WBCs, and 62 days for neutrophils. CONCLUSION: The NwSVM can be used to predict myelosuppression and clearly depicts nadir timing differences between platelets and other blood cells.