Integrated causal-predictive machine learning models for tropical cyclone epidemiology

Strategic preparedness reduces the adverse health impacts of hurricanes and tropical storms, referred to collectively as tropical cyclones (TCs), but its protective impact could be enhanced by a more comprehensive and rigorous characterization of TC epidemiology. To generate the insights and tools necessary for high-precision TC preparedness, we introduce a machine learning approach that standardizes estimation of historic TC health impacts, discovers common patterns and sources of heterogeneity in those health impacts, and enables identification of communities at highest health risk for future TCs. The model integrates (i) a causal inference component to quantify the immediate health impacts of recent historic TCs at high spatial resolution and (ii) a predictive component that captures how TC meteorological features and socioeconomic/demographic characteristics of impacted communities are associated with health impacts. We apply it to a rich data platform containing detailed historic TC exposure information and records of all-cause mortality and cardiovascular- and respiratory-related hospitalization among Medicare recipients. We report a high degree of heterogeneity in the acute health impacts of historic TCs, both within and across TCs, and, on average, substantial TC-attributable increases in respiratory hospitalizations. TC-sustained windspeeds are found to be the primary driver of mortality and respiratory risks.

A TDF-WNSP-WLFM algorithm for product recommendation based on multiple types of implicit user behavior

E-commerce platforms usually train their recommender system models to achieve personalized recommendations based on user behavior data. User behavior can be categorized into implicit and explicit feedback. Explicit feedback data have been well studied. However, the implicit feedback data still have many issues, such as the multiple types of behavior data, lack of negative feedback, and lack of the ability to express the real user preference. Targeting these problems of implicit feedback, we propose a TDF-WNSP-WLFM (time decay factor-weight of negative sample possibility-weighted latent factor model) based on the latent factor model for product recommendation. Our method mainly focuses on reconstructing the implicit rating matrix to enable the algorithm to perform better. The TDF-WNSP-WLFM algorithm is tested on two public user behavior datasets from Taobao and REES46, two big e-commerce platforms. Our algorithm compares favorably with other known collaborative filtering methods.

A sparse Bayesian factor model for the construction of gene co-expression networks from single-cell RNA sequencing count data

BACKGROUND

Gene co-expression networks (GCNs) are powerful tools that enable biologists to examine associations between genes during different biological processes. With the advancement of new technologies, such as single-cell RNA sequencing (scRNA-seq), there is a need for developing novel network methods appropriate for new types of data.

RESULTS

We present a novel sparse Bayesian factor model to explore the network structure associated with genes in scRNA-seq data. Latent factors impact the gene expression values for each cell and provide flexibility to account for common features of scRNA-seq: high proportions of zero values, increased cell-to-cell variability, and overdispersion due to abnormally large expression counts. From our model, we construct a GCN by analyzing the positive and negative associations of the factors that are shared between each pair of genes.

CONCLUSIONS

Simulation studies demonstrate that our methodology has high power in identifying gene-gene associations while maintaining a nominal false discovery rate. In real data analyses, our model identifies more known and predicted protein-protein interactions than other competing network models.

MILFM: Multiple index latent factor model based on high-dimensional features

The aim of this article is to develop a multiple-index latent factor modeling (MILFM) framework to build an accurate prediction model for clinical outcomes based on a massive number of features. We develop a three-stage estimation procedure to build the prediction model. MILFM uses an independent screening method to select a set of informative features, which may have a complex nonlinear relationship with the outcome variables. Moreover, we develop a latent factor model to project all informative predictors onto a small number of local subspaces, which lead to a few key features that capture reliable and informative covariate information. Finally, we fit the regularized empirical estimate to those key features in order to accurately predict clinical outcomes. We systematically investigate the theoretical properties of MILFM, such as risk bounds and selection consistency. Our simulation results and real data analysis show that MILFM outperforms many state-of-the-art methods in terms of prediction accuracy.

Probability-based collaborative filtering model for predicting gene-disease associations

BACKGROUND

Accurately predicting pathogenic human genes has been challenging in recent research. Considering extensive gene-disease data verified by biological experiments, we can apply computational methods to perform accurate predictions with reduced time and expenses.

METHODS

We propose a probability-based collaborative filtering model (PCFM) to predict pathogenic human genes. Several kinds of data sets, containing data of humans and data of other nonhuman species, are integrated in our model. Firstly, on the basis of a typical latent factorization model, we propose model I with an average heterogeneous regularization. Secondly, we develop modified model II with personal heterogeneous regularization to enhance the accuracy of aforementioned models. In this model, vector space similarity or Pearson correlation coefficient metrics and data on related species are also used.

RESULTS

We compared the results of PCFM with the results of four state-of-arts approaches. The results show that PCFM performs better than other advanced approaches.

CONCLUSIONS

PCFM model can be leveraged for predictions of disease genes, especially for new human genes or diseases with no known relationships.

The LZIP: A Bayesian latent factor model for correlated zero-inflated counts

Motivated by a study of molecular differences among breast cancer patients, we develop a Bayesian latent factor zero-inflated Poisson (LZIP) model for the analysis of correlated zero-inflated counts. The responses are modeled as independent zero-inflated Poisson distributions conditional on a set of subject-specific latent factors. For each outcome, we express the LZIP model as a function of two discrete random variables: the first captures the propensity to be in an underlying "at-risk" state, while the second represents the count response conditional on being at risk. The latent factors and loadings are assigned conditionally conjugate gamma priors that accommodate overdispersion and dependence among the outcomes. For posterior computation, we propose an efficient data-augmentation algorithm that relies primarily on easily sampled Gibbs steps. We conduct simulation studies to investigate both the inferential properties of the model and the computational capabilities of the proposed sampling algorithm. We apply the method to an analysis of breast cancer genomics data from The Cancer Genome Atlas.