Regression discontinuity threshold optimization.

Treatments often come with thresholds, e.g. we are given statins if our cholesterol is above a certain threshold. But which statin administration threshold maximizes our quality of life adjusted years? More generally, which threshold would optimize the average expected outcome? Regression discontinuity approaches are used to measure the local average treatment effect (LATE) and more recently also the Marginal Threshold Treatment Effect (MTTE), which shows how marginal changes in the threshold can affect the LATE. We extend this idea to define the problem of optimizing a policy threshold, i.e. selecting a threshold that optimizes the cumulative effect of the treatment on the treated. We present an estimator of the optimal threshold based on a constrained optimization framework. We show how to use machine learning (Gaussian process regression) for non-linear estimation. We also extend the estimation to a conservative threshold that is unlikely to produce harm, and we show how to include policy cost constraints. We apply these results to estimate an optimal tip-maximizing threshold for tip suggestions in taxi cabs Haggag (2014).

Out-of-Sample Tuning for Causal Discovery.

Causal discovery is continually being enriched with new algorithms for learning causal graphical probabilistic models. Each one of them requires a set of hyperparameters, creating a great number of combinations. Given that the true graph is unknown and the learning task is unsupervised, the challenge to a practitioner is how to tune these choices. We propose out-of-sample causal tuning (OCT) that aims to select an optimal combination. The method treats a causal model as a set of predictive models and uses out-of-sample protocols for supervised methods. This approach can handle general settings like latent confounders and nonlinear relationships. The method uses an information-theoretic approach to be able to generalize to mixed data types and a penalty for dense graphs to penalize for complexity. To evaluate OCT, we introduce a causal-based simulation method to create datasets that mimic the properties of real-world problems. We evaluate OCT against two other tuning approaches, based on stability and in-sample fitting. We show that OCT performs well in many experimental settings and it is an effective tuning method for causal discovery.

Using Causal Analysis for Conceptual Deep Learning Explanation.

Model explainability is essential for the creation of trustworthy Machine Learning models in healthcare. An ideal explanation resembles the decision-making process of a domain expert and is expressed using concepts or terminology that is meaningful to the clinicians. To provide such explanation, we first associate the hidden units of the classifier to clinically relevant concepts. We take advantage of radiology reports accompanying the chest X-ray images to define concepts. We discover sparse associations between concepts and hidden units using a linear sparse logistic regression. To ensure that the identified units truly influence the classifier's outcome, we adopt tools from Causal Inference literature and, more specifically, mediation analysis through counterfactual interventions. Finally, we construct a low-depth decision tree to translate all the discovered concepts into a straightforward decision rule, expressed to the radiologist. We evaluated our approach on a large chest x-ray dataset, where our model produces a global explanation consistent with clinical knowledge.

Interdisciplinary data science to advance environmental health research and improve birth outcomes.

Rates of preterm birth and low birthweight continue to rise in the United States and pose a significant public health problem. Although a variety of environmental exposures are known to contribute to these and other adverse birth outcomes, there has been a limited success in developing policies to prevent these outcomes. A better characterization of the complexities between multiple exposures and their biological responses can provide the evidence needed to inform public health policy and strengthen preventative population-level interventions. In order to achieve this, we encourage the establishment of an interdisciplinary data science framework that integrates epidemiology, toxicology and bioinformatics with biomarker-based research to better define how population-level exposures contribute to these adverse birth outcomes. The proposed interdisciplinary research framework would 1) facilitate data-driven analyses using existing data from health registries and environmental monitoring programs; 2) develop novel algorithms with the ability to predict which exposures are driving, in this case, adverse birth outcomes in the context of simultaneous exposures; and 3) refine biomarker-based research, ultimately leading to new policies and interventions to reduce the incidence of adverse birth outcomes.

Relationship Between Sleep Quality and Mood: Ecological Momentary Assessment Study.

BACKGROUND: Sleep disturbances play an important role in everyday affect and vice versa. However, the causal day-to-day interaction between sleep and mood has not been thoroughly explored, partly because of the lack of daily assessment data. Mobile phones enable us to collect ecological momentary assessment data on a daily basis in a noninvasive manner. OBJECTIVE: This study aimed to investigate the relationship between self-reported daily mood and sleep quality. METHODS: A total of 208 adult participants were recruited to report mood and sleep patterns daily via their mobile phones for 6 consecutive weeks. Participants were recruited in 4 roughly equal groups: depressed and anxious, depressed only, anxious only, and controls. The effect of daily mood on sleep quality and vice versa were assessed using mixed effects models and propensity score matching. RESULTS: All methods showed a significant effect of sleep quality on mood and vice versa. However, within individuals, the effect of sleep quality on next-day mood was much larger than the effect of previous-day mood on sleep quality. We did not find these effects to be confounded by the participants' past mood and sleep quality or other variables such as stress, physical activity, and weather conditions. CONCLUSIONS: We found that daily sleep quality and mood are related, with the effect of sleep quality on mood being significantly larger than the reverse. Correcting for participant fixed effects dramatically affected results. Causal analysis suggests that environmental factors included in the study and sleep and mood history do not mediate the relationship.

Predicting Causal Relationships from Biological Data: Applying Automated Causal Discovery on Mass Cytometry Data of Human Immune Cells.

Learning the causal relationships that define a molecular system allows us to predict how the system will respond to different interventions. Distinguishing causality from mere association typically requires randomized experiments. Methods for automated  causal discovery from limited experiments exist, but have so far rarely been tested in systems biology applications. In this work, we apply state-of-the art causal discovery methods on a large collection of public mass cytometry data sets, measuring intra-cellular signaling proteins of the human immune system and their response to several perturbations. We show how different experimental conditions can be used to facilitate causal discovery, and apply two fundamental methods that produce context-specific causal predictions. Causal predictions were reproducible across independent data sets from two different studies, but often disagree with the KEGG pathway databases. Within this context, we discuss the caveats we need to overcome for automated causal discovery to become a part of the routine data analysis in systems biology.