Infinite Continuous Feature Model for Psychiatric Comorbidity Analysis.

We aim at finding the comorbidity patterns of substance abuse, mood and personality disorders using the diagnoses from the National Epidemiologic Survey on Alcohol and Related Conditions database. To this end, we propose a novel Bayesian nonparametric latent feature model for categorical observations, based on the Indian buffet process, in which the latent variables can take values between 0 and 1. The proposed model has several interesting features for modeling psychiatric disorders. First, the latent features might be off, which allows distinguishing between the subjects who suffer a condition and those who do not. Second, the active latent features take positive values, which allows modeling the extent to which the patient has that condition. We also develop a new Markov chain Monte Carlo inference algorithm for our model that makes use of a nested expectation propagation procedure.

Handling Ill-Conditioned Omics Data With Deep Probabilistic Models.

The advent of high-throughput technologies has produced an increase in the dimensionality of omics datasets, which limits the application of machine learning methods due to the great unbalance between the number of observations and features. In this scenario, dimensionality reduction is essential to extract the relevant information within these datasets and project it in a low-dimensional space, and probabilistic latent space models are becoming popular given their capability to capture the underlying structure of the data as well as the uncertainty in the information. This article aims to provide a general classification and dimensionality reduction method based on deep latent space models that tackles two of the main problems that arise in omics datasets: the presence of missing data and the limited number of observations against the number of features. We propose a semi-supervised Bayesian latent space model that infers a low-dimensional embedding driven by the target label: the Deep Bayesian Logistic Regression (DBLR) model. During inference, the model also learns a global vector of weights that allows it to make predictions given the low-dimensional embedding of the observations. Since this kind of dataset is prone to overfitting, we introduce an additional probabilistic regularization method based on the semi-supervised nature of the model. We compared the performance of the DBLR against several state-of-the-art methods for dimensionality reduction, both in synthetic and real datasets with different data types. The proposed model provides more informative low-dimensional representations, outperforms the baseline methods in classification, and can naturally handle missing entries.

Automatic patient functionality assessment from multimodal data using deep learning techniques - Development and feasibility evaluation.

Wearable devices and mobile sensors enable the real-time collection of an abundant source of physiological and behavioural data unobtrusively. Unlike traditional in-person evaluation or ecological momentary assessment (EMA) questionnaire-based approaches, these data sources open many possibilities in remote patient monitoring. However, defining robust models is challenging due to the data's noisy and frequently missing observations. This work proposes an attention-based Long Short-Term Memory (LSTM) neural network-based pipeline for predicting mobility impairment based on WHODAS 2.0 evaluation from such digital biomarkers. Furthermore, we addressed the missing observation problem by utilising hidden Markov models and the possibility of including information from unlabelled samples via transfer learning. We validated our approach using two wearable/mobile sensor data sets collected in the wild and socio-demographic information about the patients. Our results showed that in the WHODAS 2.0 mobility impairment prediction task, the proposed pipeline outperformed a prior baseline while additionally providing interpretability with attention heatmaps. Moreover, using a much smaller cohort via task transfer learning, the same model could learn to predict generalised anxiety severity accurately based on GAD-7 scores.

Regularizing transformers with deep probabilistic layers.

Language models (LM) have grown non-stop in the last decade, from sequence-to-sequence architectures to attention-based Transformers. However, regularization is not deeply studied in those structures. In this work, we use a Gaussian Mixture Variational Autoencoder (GMVAE) as a regularizer layer. We study its advantages regarding the depth where it is placed and prove its effectiveness in several scenarios. Experimental result demonstrates that the inclusion of deep generative models within Transformer-based architectures such as BERT, RoBERTa, or XLM-R can bring more versatile models, able to generalize better and achieve improved imputation score in tasks such as SST-2 and TREC or even impute missing/noisy words with richer text.

Continuous Assessment of Function and Disability via Mobile Sensing: Real-World Data-Driven Feasibility Study.

BACKGROUND: Functional limitations are associated with poor clinical outcomes, higher mortality, and disability rates, especially in older adults. Continuous assessment of patients' functionality is important for clinical practice; however, traditional questionnaire-based assessment methods are very time-consuming and infrequently used. Mobile sensing offers a great range of sources that can assess function and disability daily. OBJECTIVE: This work aims to prove the feasibility of an interpretable machine learning pipeline for predicting function and disability based on the World Health Organization Disability Assessment Schedule (WHODAS) 2.0 outcomes of clinical outpatients, using passively collected digital biomarkers. METHODS: One-month-long behavioral time-series data consisting of physical and digital activity descriptor variables were summarized using statistical measures (minimum, maximum, mean, median, SD, and IQR), creating 64 features that were used for prediction. We then applied a sequential feature selection to each WHODAS 2.0 domain (cognition, mobility, self-care, getting along, life activities, and participation) in order to find the most descriptive features for each domain. Finally, we predicted the WHODAS 2.0 functional domain scores using linear regression using the best feature subsets. We reported the mean absolute errors and the mean absolute percentage errors over 4 folds as goodness-of-fit statistics to evaluate the model and allow for between-domain performance comparison. RESULTS: Our machine learning-based models for predicting patients' WHODAS functionality scores per domain achieved an average (across the 6 domains) mean absolute percentage error of 19.5%, varying between 14.86% (self-care domain) and 27.21% (life activities domain). We found that 5-19 features were sufficient for each domain, and the most relevant being the distance traveled, time spent at home, time spent walking, exercise time, and vehicle time. CONCLUSIONS: Our findings show the feasibility of using machine learning-based methods to assess functional health solely from passively sensed mobile data. The feature selection step provides a set of interpretable features for each domain, ensuring better explainability to the models' decisions-an important aspect in clinical practice.

Medical Data Wrangling With Sequential Variational Autoencoders.

Medical data sets are usually corrupted by noise and missing data. These missing patterns are commonly assumed to be completely random, but in medical scenarios, the reality is that these patterns occur in bursts due to sensors that are off for some time or data collected in a misaligned uneven fashion, among other causes. This paper proposes to model medical data records with heterogeneous data types and bursty missing data using sequential variational autoencoders (VAEs). In particular, we propose a new methodology, the Shi-VAE, which extends the capabilities of VAEs to sequential streams of data with missing observations. We compare our model against state-of-the-art solutions in an intensive care unit database (ICU) and a dataset of passive human monitoring. Furthermore, we find that standard error metrics such as RMSE are not conclusive enough to assess temporal models and include in our analysis the cross-correlation between the ground truth and the imputed signal. We show that Shi-VAE achieves the best performance in terms of using both metrics, with lower computational complexity than the GP-VAE model, which is the state-of-the-art method for medical records.

Predicting Emotional States Using Behavioral Markers Derived From Passively Sensed Data: Data-Driven Machine Learning Approach.

BACKGROUND: Mental health disorders affect multiple aspects of patients' lives, including mood, cognition, and behavior. eHealth and mobile health (mHealth) technologies enable rich sets of information to be collected noninvasively, representing a promising opportunity to construct behavioral markers of mental health. Combining such data with self-reported information about psychological symptoms may provide a more comprehensive and contextualized view of a patient's mental state than questionnaire data alone. However, mobile sensed data are usually noisy and incomplete, with significant amounts of missing observations. Therefore, recognizing the clinical potential of mHealth tools depends critically on developing methods to cope with such data issues. OBJECTIVE: This study aims to present a machine learning-based approach for emotional state prediction that uses passively collected data from mobile phones and wearable devices and self-reported emotions. The proposed methods must cope with high-dimensional and heterogeneous time-series data with a large percentage of missing observations. METHODS: Passively sensed behavior and self-reported emotional state data from a cohort of 943 individuals (outpatients recruited from community clinics) were available for analysis. All patients had at least 30 days' worth of naturally occurring behavior observations, including information about physical activity, geolocation, sleep, and smartphone app use. These regularly sampled but frequently missing and heterogeneous time series were analyzed with the following probabilistic latent variable models for data averaging and feature extraction: mixture model (MM) and hidden Markov model (HMM). The extracted features were then combined with a classifier to predict emotional state. A variety of classical machine learning methods and recurrent neural networks were compared. Finally, a personalized Bayesian model was proposed to improve performance by considering the individual differences in the data and applying a different classifier bias term for each patient. RESULTS: Probabilistic generative models proved to be good preprocessing and feature extractor tools for data with large percentages of missing observations. Models that took into account the posterior probabilities of the MM and HMM latent states outperformed those that did not by more than 20%, suggesting that the underlying behavioral patterns identified were meaningful for individuals' overall emotional state. The best performing generalized models achieved a 0.81 area under the curve of the receiver operating characteristic and 0.71 area under the precision-recall curve when predicting self-reported emotional valence from behavior in held-out test data. Moreover, the proposed personalized models demonstrated that accounting for individual differences through a simple hierarchical model can substantially improve emotional state prediction performance without relying on previous days' data. CONCLUSIONS: These findings demonstrate the feasibility of designing machine learning models for predicting emotional states from mobile sensing data capable of dealing with heterogeneous data with large numbers of missing observations. Such models may represent valuable tools for clinicians to monitor patients' mood states.

Deep Sequential Models for Suicidal Ideation From Multiple Source Data.

This paper presents a novel method for predicting suicidal ideation from electronic health records (EHR) and ecological momentary assessment (EMA) data using deep sequential models. Both EHR longitudinal data and EMA question forms are defined by asynchronous, variable length, randomly sampled data sequences. In our method, we model each of them with a recurrent neural network, and both sequences are aligned by concatenating the hidden state of each of them using temporal marks. Furthermore, we incorporate attention schemes to improve performance in long sequences and time-independent pre-trained schemes to cope with very short sequences. Using a database of 1023 patients, our experimental results show that the addition of EMA records boosts the system recall to predict the suicidal ideation diagnosis from 48.13% obtained exclusively from EHR-based state-of-the-art methods to 67.78%. Additionally, our method provides interpretability through the t-distributed stochastic neighbor embedding (t-SNE) representation of the latent space. Furthermore, the most relevant input features are identified and interpreted medically.