Deep learning methods in metagenomics: a review.

The ever-decreasing cost of sequencing and the growing potential applications of metagenomics have led to an unprecedented surge in data generation. One of the most prevalent applications of metagenomics is the study of microbial environments, such as the human gut. The gut microbiome plays a crucial role in human health, providing vital information for patient diagnosis and prognosis. However, analysing metagenomic data remains challenging due to several factors, including reference catalogues, sparsity and compositionality. Deep learning (DL) enables novel and promising approaches that complement state-of-the-art microbiome pipelines. DL-based methods can address almost all aspects of microbiome analysis, including novel pathogen detection, sequence classification, patient stratification and disease prediction. Beyond generating predictive models, a key aspect of these methods is also their interpretability. This article reviews DL approaches in metagenomics, including convolutional networks, autoencoders and attention-based models. These methods aggregate contextualized data and pave the way for improved patient care and a better understanding of the microbiome's key role in our health.

Functional alterations and predictive capacity of gut microbiome in type 2 diabetes.

The gut microbiome plays a significant role in the development of Type 2 Diabetes Mellitus (T2DM), but the functional mechanisms behind this association merit deeper investigation. Here, we used the nanopore sequencing technology for metagenomic analyses to compare the gut microbiome of individuals with T2DM from the United Arab Emirates (n = 40) with that of control (n = 44). DMM enterotyping of the cohort resulted concordantly with previous results, in three dominant groups Bacteroides (K1), Firmicutes (K2), and Prevotella (K3) lineages. The diversity analysis revealed a high level of diversity in the Firmicutes group (K2) both in terms of species richness and evenness (Wilcoxon rank-sum test, p value < 0.05 vs. K1 and K3 groups), consistent with the Ruminococcus enterotype described in Western populations. Additionally, functional enrichment analyses of KEGG modules showed significant differences in abundance between individuals with T2DM and controls (FDR < 0.05). These differences include modules associated with the degradation of amino acids, such as arginine, the degradation of urea as well as those associated with homoacetogenesis. Prediction analysis with the Predomics approach suggested potential biomarkers for T2DM, including a balance between a depletion of Enterococcus faecium and Blautia lineages with an enrichment of Absiella spp or Eubacterium limosum in T2DM individuals, highlighting the potential of metagenomic analysis in predicting predisposition to diabetic cardiomyopathy in T2DM patients.

Evidence of a causal and modifiable relationship between kidney function and circulating trimethylamine N-oxide.

The host-microbiota co-metabolite trimethylamine N-oxide (TMAO) is linked to increased cardiovascular risk but how its circulating levels are regulated remains unclear. We applied "explainable" machine learning, univariate, multivariate and mediation analyses of fasting plasma TMAO concentration and a multitude of phenotypes in 1,741 adult Europeans of the MetaCardis study. Here we show that next to age, kidney function is the primary variable predicting circulating TMAO, with microbiota composition and diet playing minor, albeit significant, roles. Mediation analysis suggests a causal relationship between TMAO and kidney function that we corroborate in preclinical models where TMAO exposure increases kidney scarring. Consistent with our findings, patients receiving glucose-lowering drugs with reno-protective properties have significantly lower circulating TMAO when compared to propensity-score matched control individuals. Our analyses uncover a bidirectional relationship between kidney function and TMAO that can potentially be modified by reno-protective anti-diabetic drugs and suggest a clinically actionable intervention for decreasing TMAO-associated excess cardiovascular risk.

Automatic digitization of paper electrocardiograms - A systematic review.

The digitization of electrocardiogram paper records is an essential step to preserve and analyze cardiac data. This digitization process is not flawless as it involves several challenges, such as skew correction, binarization, and signal extraction. Various approaches have been proposed to address these challenges and recent studies have introduced innovative solutions, such as deep learning models and automation processes. Although existing approaches have shown promising results, there is a lack of common databases and metrics where authors could evaluate and compare their methods. Furthermore, the limited accessibility of code or software hinders the comparison process. Overall, while digitization of paper ECG recordings is important in advancing cardiology research, additional efforts are needed to standardize the evaluation process while improving code accessibility. This article provides a systematic review of this process.

LiSA: an assisted literature search pipeline for detecting serious adverse drug events with deep learning.

INTRODUCTION: Detecting safety signals attributed to a drug in scientific literature is a fundamental issue in pharmacovigilance. The constant increase in the volume of publications requires the automation of this tedious task, in order to find and extract relevant articles from the pack. This task is critical, as serious Adverse Drug Reactions (ADRs) still account for a large number of hospital admissions each year. OBJECTIVES: The aim of this study is to develop an augmented intelligence methodology for automatically identifying relevant publications mentioning an established link between a Drug and a Serious Adverse Event, according to the European Medicines Agency (EMA) definition of seriousness. METHODS: The proposed pipeline, called LiSA (for Literature Search Application), is based on three independent deep learning models supporting a precise detection of safety signals in the biomedical literature. By combining a Bidirectional Encoder Representations from Transformers (BERT) algorithms and a modular architecture, the pipeline achieves a precision of 0.81 and a recall of 0.89 at sentences level in articles extracted from PubMed (either abstract or full-text). We also measured that by using LiSA, a medical reviewer increases by a factor of 2.5 the number of relevant documents it can collect and evaluate compared to a simple keyword search. In the interest of re-usability, emphasis was placed on building a modular pipeline allowing the insertion of other NLP modules to enrich the results provided by the system, and extend it to other use cases. In addition, a lightweight visualization tool was developed to analyze and monitor safety signal results. CONCLUSIONS: Overall, the generic pipeline and the visualization tool proposed in this article allows for efficient and accurate monitoring of serious adverse drug reactions from the literature and can easily be adapted to similar pharmacovigilance use cases. To facilitate reproducibility and benefit other research studies, we also shared a first benchmark dataset for Serious Adverse Drug Events detection.

The human gut microbiota contributes to type-2 diabetes non-resolution 5-years after Roux-en-Y gastric bypass.

Roux-en-Y gastric bypass (RYGB) is efficient at inducing drastic albeit variable weight loss and type-2 diabetes (T2D) improvements in patients with severe obesity and T2D. We hypothesized a causal implication of the gut microbiota (GM) in these metabolic benefits, as RYGB is known to deeply impact its composition. In a cohort of 100 patients with baseline T2D who underwent RYGB and were followed for 5-years, we used a hierarchical clustering approach to stratify subjects based on the severity of their T2D (Severe vs Mild) throughout the follow-up. We identified via nanopore-based GM sequencing that the more severe cases of unresolved T2D were associated with a major increase of the class Bacteroidia, including 12 species comprising Phocaeicola dorei, Bacteroides fragilis, and Bacteroides caecimuris. A key observation is that patients who underwent major metabolic improvements do not harbor this enrichment in Bacteroidia, as those who presented mild cases of T2D at all times. In a separate group of 36 patients with similar baseline clinical characteristics and preoperative GM sequencing, we showed that this increase in Bacteroidia was already present at baseline in the most severe cases of T2D. To explore the causal relationship linking this enrichment in Bacteroidia and metabolic alterations, we selected 13 patients across T2D severity clusters at 5-years and performed fecal matter transplants in mice. Our results show that 14 weeks after the transplantations, mice colonized with the GM of Severe donors have impaired glucose tolerance and insulin sensitivity as compared to Mild-recipients, all in the absence of any difference in body weight and composition. GM sequencing of the recipient animals revealed that the hallmark T2D-severity associated bacterial features were transferred and were associated with the animals' metabolic alterations. Therefore, our results further establish the GM as a key contributor to long-term glucose metabolism improvements (or lack thereof) after RYGB.

Qluster: An easy-to-implement generic workflow for robust clustering of health data.

The exploration of heath data by clustering algorithms allows to better describe the populations of interest by seeking the sub-profiles that compose it. This therefore reinforces medical knowledge, whether it is about a disease or a targeted population in real life. Nevertheless, contrary to the so-called conventional biostatistical methods where numerous guidelines exist, the standardization of data science approaches in clinical research remains a little discussed subject. This results in a significant variability in the execution of data science projects, whether in terms of algorithms used, reliability and credibility of the designed approach. Taking the path of parsimonious and judicious choice of both algorithms and implementations at each stage, this article proposes Qluster, a practical workflow for performing clustering tasks. Indeed, this workflow makes a compromise between (1) genericity of applications (e.g. usable on small or big data, on continuous, categorical or mixed variables, on database of high-dimensionality or not), (2) ease of implementation (need for few packages, few algorithms, few parameters, ...), and (3) robustness (e.g. use of proven algorithms and robust packages, evaluation of the stability of clusters, management of noise and multicollinearity). This workflow can be easily automated and/or routinely applied on a wide range of clustering projects. It can be useful both for data scientists with little experience in the field to make data clustering easier and more robust, and for more experienced data scientists who are looking for a straightforward and reliable solution to routinely perform preliminary data mining. A synthesis of the literature on data clustering as well as the scientific rationale supporting the proposed workflow is also provided. Finally, a detailed application of the workflow on a concrete use case is provided, along with a practical discussion for data scientists. An implementation on the Dataiku platform is available upon request to the authors.

Combinatorial, additive and dose-dependent drug-microbiome associations.

During the transition from a healthy state to cardiometabolic disease, patients become heavily medicated, which leads to an increasingly aberrant gut microbiome and serum metabolome, and complicates biomarker discovery1-5. Here, through integrated multi-omics analyses of 2,173 European residents from the MetaCardis cohort, we show that the explanatory power of drugs for the variability in both host and gut microbiome features exceeds that of disease. We quantify inferred effects of single medications, their combinations as well as additive effects, and show that the latter shift the metabolome and microbiome towards a healthier state, exemplified in synergistic reduction in serum atherogenic lipoproteins by statins combined with aspirin, or enrichment of intestinal Roseburia by diuretic agents combined with beta-blockers. Several antibiotics exhibit a quantitative relationship between the number of courses prescribed and progression towards a microbiome state that is associated with the severity of cardiometabolic disease. We also report a relationship between cardiometabolic drug dosage, improvement in clinical markers and microbiome composition, supporting direct drug effects. Taken together, our computational framework and resulting resources enable the disentanglement of the effects of drugs and disease on host and microbiome features in multimedicated individuals. Furthermore, the robust signatures identified using our framework provide new hypotheses for drug-host-microbiome interactions in cardiometabolic disease.

Exploring Semi-Quantitative Metagenomic Studies Using Oxford Nanopore Sequencing: A Computational and Experimental Protocol.

The gut microbiome plays a major role in chronic diseases, of which several are characterized by an altered composition and diversity of bacterial communities. Large-scale sequencing projects allowed for characterizing the perturbations of these communities. However, translating these discoveries into clinical applications remains a challenge. To facilitate routine implementation of microbiome profiling in clinical settings, portable, real-time, and low-cost sequencing technologies are needed. Here, we propose a computational and experimental protocol for whole-genome semi-quantitative metagenomic studies of human gut microbiome with Oxford Nanopore sequencing technology (ONT) that could be applied to other microbial ecosystems. We developed a bioinformatics protocol to analyze ONT sequences taxonomically and functionally and optimized preanalytic protocols, including stool collection and DNA extraction methods to maximize read length. This is a critical parameter for the sequence alignment and classification. Our protocol was evaluated using simulations of metagenomic communities, which reflect naturally occurring compositional variations. Next, we validated both protocols using stool samples from a bariatric surgery cohort, sequenced with ONT, Illumina, and SOLiD technologies. Results revealed similar diversity and microbial composition profiles. This protocol can be implemented in a clinical or research setting, bringing rapid personalized whole-genome profiling of target microbiome species.

Deep learning analysis of electrocardiogram for risk prediction of drug-induced arrhythmias and diagnosis of long QT syndrome.

AIMS: Congenital long-QT syndromes (cLQTS) or drug-induced long-QT syndromes (diLQTS) can cause torsade de pointes (TdP), a life-threatening ventricular arrhythmia. The current strategy for the identification of drugs at the high risk of TdP relies on measuring the QT interval corrected for heart rate (QTc) on the electrocardiogram (ECG). However, QTc has a low positive predictive value. METHODS AND RESULTS: We used convolutional neural network (CNN) models to quantify ECG alterations induced by sotalol, an IKr blocker associated with TdP, aiming to provide new tools (CNN models) to enhance the prediction of drug-induced TdP (diTdP) and diagnosis of cLQTS. Tested CNN models used single or multiple 10-s recordings/patient using 8 leads or single leads in various cohorts: 1029 healthy subjects before and after sotalol intake (n = 14 135 ECGs); 487 cLQTS patients (n = 1083 ECGs: 560 type 1, 456 type 2, 67 type 3); and 48 patients with diTdP (n = 1105 ECGs, with 147 obtained within 48 h of a diTdP episode). CNN models outperformed models using QTc to identify exposure to sotalol [area under the receiver operating characteristic curve (ROC-AUC) = 0.98 vs. 0.72, P ≤ 0.001]. CNN models had higher ROC-AUC using multiple vs. single 10-s ECG (P ≤ 0.001). Performances were comparable for 8-lead vs. single-lead models. CNN models predicting sotalol exposure also accurately detected the presence and type of cLQTS vs. healthy controls, particularly for cLQT2 (AUC-ROC = 0.9) and were greatest shortly after a diTdP event and declining over time (P ≤ 0.001), after controlling for QTc and intake of culprit drugs. ECG segment analysis identified the J-Tpeak interval as the best discriminator of sotalol intake. CONCLUSION: CNN models applied to ECGs outperform QTc measurements to identify exposure to drugs altering the QT interval, congenital LQTS, and are greatest shortly after a diTdP episode.

Protein supplementation during an energy-restricted diet induces visceral fat loss and gut microbiota amino acid metabolism activation: a randomized trial.

Interactions between diet and gut microbiota are critical regulators of energy metabolism. The effects of fibre intake have been deeply studied but little is known about the impact of proteins. Here, we investigated the effects of high protein supplementation (Investigational Product, IP) in a double blind, randomised placebo-controled intervention study (NCT01755104) where 107 participants received the IP or an isocaloric normoproteic comparator (CP) alongside a mild caloric restriction. Gut microbiota profiles were explored in a patient subset (n = 53) using shotgun metagenomic sequencing. Visceral fat decreased in both groups (IP group: - 20.8 ± 23.2 cm2; CP group: - 14.5 ± 24.3 cm2) with a greater reduction (p < 0.05) with the IP supplementation in the Per Protocol population. Microbial diversity increased in individuals with a baseline low gene count (p < 0.05). The decrease in weight, fat mass and visceral fat mass significantly correlated with the increase in microbial diversity (p < 0.05). Protein supplementation had little effects on bacteria composition but major differences were seen at functional level. Protein supplementation stimulated bacterial amino acid metabolism (90% amino-acid synthesis functions enriched with IP versus 13% in CP group (p < 0.01)). Protein supplementation alongside a mild energy restriction induces visceral fat mass loss and an activation of gut microbiota amino-acid metabolism.Clinical trial registration: NCT01755104 (24/12/2012). https://clinicaltrials.gov/ct2/show/record/NCT01755104?term=NCT01755104&draw=2&rank=1 .

Characterization of the Gut Microbiota in Individuals with Overweight or Obesity during a Real-World Weight Loss Dietary Program: A Focus on the Bacteroides 2 Enterotype.

BACKGROUND: Dietary intervention is a cornerstone of weight loss therapies. In obesity, a dysbiotic gut microbiota (GM) is characterized by high levels of Bacteroides lineages and low diversity. We examined the GM composition changes, including the Bacteroides 2 enterotype (Bact2), in a real-world weight loss study in subjects following a high-protein hypocaloric diet with or without a live microorganisms (LMP) supplement. METHOD: 263 volunteers were part of this real-world weight loss program. The first phase was a high-protein low-carbohydrate calorie restriction diet with or without LMP supplements. Fecal samples were obtained at baseline and after 10% weight loss for 163 subjects. Metagenomic profiling was obtained by shotgun sequencing. RESULTS: At baseline, the Bact2 enterotype was more prevalent in subjects with aggravated obesity and metabolic alterations. After weight loss, diversity increased and Bact2 prevalence decreased in subjects with lower GM diversity at baseline, notably in LMP consumers. Significant increases in Akkermansia muciniphila and Parabacteroides distasonis and significant decreases of Eubacterium rectale, Streptococcus thermophilus and Bifidobacterial lineages were observed after weight loss. CONCLUSIONS: Baseline microbiome composition is associated with differential changes in GM diversity and Bact2 enterotype prevalence after weight loss. Examining these signatures could drive future personalized nutrition efforts towards more favorable microbiome compositions.

From correlation to causality: the case of Subdoligranulum.

Gut microbes are considered as major factors contributing to human health. Nowadays, the vast majority of the data available in the literature are mostly exhibiting negative or positive correlations between specific bacteria and metabolic parameters. From these observations, putative detrimental or beneficial effects are then inferred. Akkermansia muciniphila is one of the unique examples for which the correlations with health benefits have been causally validated in vivo in rodents and humans. In this study, based on available metagenomic data in overweight/obese population and clinical variables that we obtained from two cohorts of individuals (n = 108) we identified several metagenomic species (MGS) strongly associated with A. muciniphila with one standing out: Subdoligranulum. By analyzing both qPCR and shotgun metagenomic data, we discovered that the abundance of Subdoligranulum was correlated positively with microbial richness and HDL-cholesterol levels and negatively correlated with fat mass, adipocyte diameter, insulin resistance, levels of leptin, insulin, CRP, and IL6 in humans. Therefore, to further explore whether these strong correlations could be translated into causation, we investigated the effects of the unique cultivated strain of Subdoligranulum (Subdoligranulum variabile DSM 15176 T) in obese and diabetic mice as a proof-of-concept. Strikingly, there were no significant difference in any of the hallmarks of obesity and diabetes measured (e.g., body weight gain, fat mass gain, glucose tolerance, liver weight, plasma lipids) at the end of the 8 weeks of treatment. Therefore, the absence of effect following the supplementation with S. variabile indicates that increasing the intestinal abundance of this bacterium is not translated into beneficial effects in mice. In conclusion, we demonstrated that despite the fact that numerous strong correlations exist between a given bacteria and health, proof-of-concept experiments are required to be further validated or not in vivo. Hence, an urgent need for causality studies is warranted to move from human observations to preclinical validations.

Imidazole propionate is increased in diabetes and associated with dietary patterns and altered microbial ecology.

Microbiota-host-diet interactions contribute to the development of metabolic diseases. Imidazole propionate is a novel microbially produced metabolite from histidine, which impairs glucose metabolism. Here, we show that subjects with prediabetes and diabetes in the MetaCardis cohort from three European countries have elevated serum imidazole propionate levels. Furthermore, imidazole propionate levels were increased in subjects with low bacterial gene richness and Bacteroides 2 enterotype, which have previously been associated with obesity. The Bacteroides 2 enterotype was also associated with increased abundance of the genes involved in imidazole propionate biosynthesis from dietary histidine. Since patients and controls did not differ in their histidine dietary intake, the elevated levels of imidazole propionate in type 2 diabetes likely reflects altered microbial metabolism of histidine, rather than histidine intake per se. Thus the microbiota may contribute to type 2 diabetes by generating imidazole propionate that can modulate host inflammation and metabolism.

Statin therapy is associated with lower prevalence of gut microbiota dysbiosis.

Microbiome community typing analyses have recently identified the Bacteroides2 (Bact2) enterotype, an intestinal microbiota configuration that is associated with systemic inflammation and has a high prevalence in loose stools in humans1,2. Bact2 is characterized by a high proportion of Bacteroides, a low proportion of Faecalibacterium and low microbial cell densities1,2, and its prevalence varies from 13% in a general population cohort to as high as 78% in patients with inflammatory bowel disease2. Reported changes in stool consistency3 and inflammation status4 during the progression towards obesity and metabolic comorbidities led us to propose that these developments might similarly correlate with an increased prevalence of the potentially dysbiotic Bact2 enterotype. Here, by exploring obesity-associated microbiota alterations in the quantitative faecal metagenomes of the cross-sectional MetaCardis Body Mass Index Spectrum cohort (n = 888), we identify statin therapy as a key covariate of microbiome diversification. By focusing on a subcohort of participants that are not medicated with statins, we find that the prevalence of Bact2 correlates with body mass index, increasing from 3.90% in lean or overweight participants to 17.73% in obese participants. Systemic inflammation levels in Bact2-enterotyped individuals are higher than predicted on the basis of their obesity status, indicative of Bact2 as a dysbiotic microbiome constellation. We also observe that obesity-associated microbiota dysbiosis is negatively associated with statin treatment, resulting in a lower Bact2 prevalence of 5.88% in statin-medicated obese participants. This finding is validated in both the accompanying MetaCardis cardiovascular disease dataset (n = 282) and the independent Flemish Gut Flora Project population cohort (n = 2,345). The potential benefits of statins in this context will require further evaluation in a prospective clinical trial to ascertain whether the effect is reproducible in a randomized population and before considering their application as microbiota-modulating therapeutics.

Interpretable and accurate prediction models for metagenomics data.

BACKGROUND: Microbiome biomarker discovery for patient diagnosis, prognosis, and risk evaluation is attracting broad interest. Selected groups of microbial features provide signatures that characterize host disease states such as cancer or cardio-metabolic diseases. Yet, the current predictive models stemming from machine learning still behave as black boxes and seldom generalize well. Their interpretation is challenging for physicians and biologists, which makes them difficult to trust and use routinely in the physician-patient decision-making process. Novel methods that provide interpretability and biological insight are needed. Here, we introduce "predomics", an original machine learning approach inspired by microbial ecosystem interactions that is tailored for metagenomics data. It discovers accurate predictive signatures and provides unprecedented interpretability. The decision provided by the predictive model is based on a simple, yet powerful score computed by adding, subtracting, or dividing cumulative abundance of microbiome measurements. RESULTS: Tested on >100 datasets, we demonstrate that predomics models are simple and highly interpretable. Even with such simplicity, they are at least as accurate as state-of-the-art methods. The family of best models, discovered during the learning process, offers the ability to distil biological information and to decipher the predictability signatures of the studied condition. In a proof-of-concept experiment, we successfully predicted body corpulence and metabolic improvement after bariatric surgery using pre-surgery microbiome data. CONCLUSIONS: Predomics is a new algorithm that helps in providing reliable and trustworthy diagnostic decisions in the microbiome field. Predomics is in accord with societal and legal requirements that plead for an explainable artificial intelligence approach in the medical field.

Gut Microbiota Profile of Obese Diabetic Women Submitted to Roux-en-Y Gastric Bypass and Its Association with Food Intake and Postoperative Diabetes Remission.

Gut microbiota composition is influenced by environmental factors and has been shown to impact body metabolism. OBJECTIVE: To assess the gut microbiota profile before and after Roux-en-Y gastric bypass (RYGB) and the correlation with food intake and postoperative type 2 diabetes remission (T2Dr). DESIGN: Gut microbiota profile from obese diabetic women was evaluated before (n = 25) and 3 (n = 20) and 12 months (n = 14) after RYGB, using MiSeq Illumina-based V4 bacterial 16S rRNA gene profiling. Data on food intake (7-day record) and T2Dr (American Diabetes Association (ADA) criteria) were recorded. RESULTS: Preoperatively, the abundance of five bacteria genera differed between patients with (57%) and without T2Dr (p < 0.050). Preoperative gut bacteria genus signature was able to predict the T2Dr status with 0.94 accuracy ROC curve (receiver operating characteristic curve). Postoperatively (vs. preoperative), the relative abundance of some gut bacteria genera changed, the gut microbial richness increased, and the Firmicutes to Bacteroidetes ratio (rFB) decreased (p < 0.05) regardless of T2Dr. Richness levels was correlated with dietary profile pre and postoperatively, mainly displaying positive and inverse correlations with fiber and lipid intakes, respectively (p < 0.05). CONCLUSIONS: Gut microbiota profile was influenced by RYGB and correlated with diet and T2Dr preoperatively, suggesting the possibility to assess its composition to predict postoperative T2Dr.

Q-Finder: An Algorithm for Credible Subgroup Discovery in Clinical Data Analysis - An Application to the International Diabetes Management Practice Study.

Addressing the heterogeneity of both the outcome of a disease and the treatment response to an intervention is a mandatory pathway for regulatory approval of medicines. In randomized clinical trials (RCTs), confirmatory subgroup analyses focus on the assessment of drugs in predefined subgroups, while exploratory ones allow a posteriori the identification of subsets of patients who respond differently. Within the latter area, subgroup discovery (SD) data mining approach is widely used-particularly in precision medicine-to evaluate treatment effect across different groups of patients from various data sources (be it from clinical trials or real-world data). However, both the limited consideration by standard SD algorithms of recommended criteria to define credible subgroups and the lack of statistical power of the findings after correcting for multiple testing hinder the generation of hypothesis and their acceptance by healthcare authorities and practitioners. In this paper, we present the Q-Finder algorithm that aims to generate statistically credible subgroups to answer clinical questions, such as finding drivers of natural disease progression or treatment response. It combines an exhaustive search with a cascade of filters based on metrics assessing key credibility criteria, including relative risk reduction assessment, adjustment on confounding factors, individual feature's contribution to the subgroup's effect, interaction tests for assessing between-subgroup treatment effect interactions and tests adjustment (multiple testing). This allows Q-Finder to directly target and assess subgroups on recommended credibility criteria. The top-k credible subgroups are then selected, while accounting for subgroups' diversity and, possibly, clinical relevance. Those subgroups are tested on independent data to assess their consistency across databases, while preserving statistical power by limiting the number of tests. To illustrate this algorithm, we applied it on the database of the International Diabetes Management Practice Study (IDMPS) to better understand the drivers of improved glycemic control and rate of episodes of hypoglycemia in type 2 diabetics patients. We compared Q-Finder with state-of-the-art approaches from both Subgroup Identification and Knowledge Discovery in Databases literature. The results demonstrate its ability to identify and support a short list of highly credible and diverse data-driven subgroups for both prognostic and predictive tasks.

Using Unlabeled Data to Discover Bivariate Causality with Deep Restricted Boltzmann Machines.

An important question in microbiology is whether treatment causes changes in gut flora, and whether it also affects metabolism. The reconstruction of causal relations purely from non-temporal observational data is challenging. We address the problem of causal inference in a bivariate case, where the joint distribution of two variables is observed. We consider, in particular, data on discrete domains. The state-of-the-art causal inference methods for continuous data suffer from high computational complexity. Some modern approaches are not suitable for categorical data, and others need to estimate and fix multiple hyper-parameters. In this contribution, we introduce a novel method of causal inference which is based on the widely used assumption that if X causes Y, then P(X) and P(Y|X) are independent. We propose to explore a semi-supervised approach where P(Y|X) and P(X) are estimated from labeled and unlabeled data respectively, whereas the marginal probability is estimated potentially from much more (cheap unlabeled) data than the conditional distribution. We validate the proposed method on the standard cause-effect pairs. We illustrate by experiments on several benchmarks of biological network reconstruction that the proposed approach is very competitive in terms of computational time and accuracy compared to the state-of-the-art methods. Finally, we apply the proposed method to an original medical task where we study whether drugs confound human metagenome.