Causal Features Research Articles

Multi-label causal feature selection based on neighbourhood mutual information

Multi-label causal feature selection based on neighbourhood mutual information

High Dimensional Multiomics Reveals Unique Characteristics of Early Plasma Administration in Polytrauma Patients With TBI.

The authors sought to identify causal factors that explain the selective benefit of prehospital administration of thawed plasma (TP) in traumatic brain injury (TBI) patients using mediation analysis of a multiomic database. The Prehospital Air Medical Plasma (PAMPer) Trial showed that patients with TBI and a pronounced systemic response to injury [defined as endotype 2 (E2)], have a survival benefit from prehospital administration of TP. An interrogation of high dimensional proteomics, lipidomics and metabolomics previously demonstrated unique patterns in circulating biomarkers in patients receiving prehospital TP, suggesting that a deeper analysis could reveal causal features specific to TBI patients. A novel proteomic database (SomaLogic Inc., aptamer-based assay, 7K platform) was generated using admission blood samples from a subset of patients (n=149) from the PAMPer Trial. This proteomic dataset was combined with previously reported metabolomic and lipidomic datasets from these same patients. A 2-step analysis was performed to identify factors that promote survival in E2-TBI patients who had received early TP. First, features were selected using both linear and multivariate-latent-factor regression analyses. Then, the selected features were entered into the causal mediation analysis. Causal mediation analysis of observable features identified 16 proteins and 41 lipids with a high proportion of mediated effect (>50%) to explain the survival benefit of early TP in E2-TBI patients. The multivariate latent-factor regression analyses also uncovered 5 latent clusters of features with a proportion effect >30%, many in common with the observable features. Among the observable and latent features were protease inhibitors known to inhibit activated protein C and block fibrinolysis (SERPINA5 and CPB2), a clotting factor (factor XI), as well as proteins involved in lipid transport and metabolism (APOE3 and sPLA(2)-XIIA). These findings suggest that severely injured patients with TBI process exogenous plasma differently than those without TBI. The beneficial effects of early TP in E2-TBI patients may be the result of improved blood clotting and the effect of brain protective factors independent of coagulation.

Multilabel causal variable discovery in multisource

AbstractMultilabel causal feature selection, as a well‐known and effective approach in dealing with high‐dimensional multilabel data, is a popular topic. Amount of causal feature selection algorithms have achieved a great deal of success in classification and prediction tasks. However, the descriptive information of data is collected from different data sources in many practical applications. While few researches focus on the causal variable discovery in multisource environments due to the complex causal relationships. To address these problems, we propose a causal feature selection framework in multisource environments to solve the above problems. Firstly, we mine the causal mechanism with respect to the class attribute under the assumption that only a single data source is included. Secondly, by utilizing the concept of causal invariance in causal inference, we formulate the problem of causal feature selection with multiple data sources as a search problem for an invariant set across data sources. In addition, we give the upper and lower bounds of the causal invariant set. Finally, we design a novel multisource multilabel causal feature selection (MMCFS) algorithm. To verify the effectiveness of the proposed algorithm, we compare it with 12 feature selection methods on synthetic datasets. Experiment results show that the classification performance of MMCFS achieves highly competitive performance against other comparing algorithms.

Interpretable instance disease prediction based on causal feature selection and effect analysis

BackgroundIn the big wave of artificial intelligence sweeping the world, machine learning has made great achievements in healthcare in the past few years, however, these methods are only based on correlation, not causation. The particularities of the healthcare determines that the research method must comply with the causality norm, otherwise the wrong intervention measures may bring the patients a lifetime of misfortune.MethodsWe propose a two-stage prediction method (instance feature selection prediction and causal effect analysis) for instance disease prediction. Feature selection is based on the counterfactual and uses the reinforcement learning framework to design an interpretable qualitative instance feature selection prediction. The model is composed of three neural networks (counterfactual prediction network, fact prediction network and counterfactual feature selection network), and the actor-critical method is used to train the network. Then we take the counterfactual prediction network as a structured causal model and improve the neural network attribution algorithm based on gradient integration to quantitatively calculate the causal effect of selection features on the output results.ResultsThe results of our experiments on synthetic data, open source data and real medical data show that our proposed method can provide qualitative and quantitative causal explanations for the model while giving prediction results.ConclusionsThe experimental results demonstrate that causality can further explore more essential relationships between variables and the prediction method based on causal feature selection and effect analysis can build a more reliable disease prediction model.

Stable soft sensor modeling based on causality analysis

Data-driven soft sensors, aiming to estimate and predict hard-to-measure quality variables using easy-to-measure process variables, have now become the key foundation for monitoring the stable and safe operation of industrial processes. However, traditional machine-learning methods usually make an assumption that training data and test data share the same probability distribution or the probability distribution of test data is known, which is impractical in the fact that test data come from multi-unknown operating modes. Based on causality analysis and stable learning, soft sensors for stable prediction, namely stable soft sensors, are proposed in this paper. To address this problem, three stable soft sensor frameworks based on causal variables, unsupervised causal features, and supervised causal features are designed. By introducing causality in soft sensor modeling, the interpretability is enhanced and the prediction results in different operating modes get stable. The effectiveness of the proposed method is shown through case studies in the benchmark Tennessee Eastman process.

Survival Patterns of Human Prion Diseases in Spain, 1998-2018: Clinical Phenotypes and Etiological Clues.

BackgroundHuman transmissible spongiform encephalopathies (TSEs) are a group of fatal neurodegenerative disorders of short duration. There are few studies on TSE survival. This study sought to analyze the survival and related factors of a TSE patient cohort, based on a nationwide surveillance system in Spain.MethodsSurvival analyses were performed on 1,530 cases diagnosed across the period 1998–2018 in Spain. We calculated median survival times and plotted survival curves using the Kaplan–Meier method for all cases and for sporadic TSE (sTSE) and genetic TSE (gTSE). Crude and adjusted Cox proportional hazard models were used to identify variables associated with shorter survival.FindingsMedian age at onset decreased from the sporadic forms to gTSE and, lastly, to acquired TSE. Overall median and interquartile range (IQR) survival time was 5.2 (IQR, 3.0–11.7) months and 4.9 (IQR, 2.8–10.8) months in sporadic cases and 9 (IQR, 4.9 to over 12) months in genetic cases, p < 0.001. Male sex, older age at onset, presence of 14-3-3 protein, typical MRI, and MM and VV polymorphisms at codon 129 were associated with shorter survival. gTSE showed higher survival in crude comparisons but not after adjustment.InterpretationTSE survival in Spain replicates both the magnitude of that shown and the TSE entity-specific population patterns observed in Western countries but differs from features described in Asian populations, such as the Japanese. The reduction in differences in survival between gTSE and sTSE on adjusting for covariates and international patterns might support the view that gTSE and sTSE share causal and pathophysiological features.

SCAN: A shared causal attention network for adverse drug reactions detection in tweets

Twitter is a popular social media site on which people post millions of Tweets every day. As patients often share their experiences with drugs on Twitter, Tweets can also be considered as a rich alternative source of adverse drug reaction (ADR)-related information. This information can be useful for health authorities and drug manufacturing companies to monitor the post-marketing effectiveness of drugs. However, the automatic detection of ADRs in Tweets is challenging, as Tweets are informal and prone to grammatical errors. The existing approaches to automatically detecting ADRs do not consider the cause-effect relationships between a drug and an ADR. In this paper, we propose a novel shared causal attention network that exploits such cause-effect relationships to detect ADRs in Tweets. In our approach, we split a Tweet into the prefix, midfix, and postfix segments based on the position of the drug name in the Tweet and separately extract causal features from the segments. We then share these separate causal features with both word and parts-of-speech features, and apply the multi-head self-attention mechanism. We run extensive experiments on three publicly available benchmark datasets to illustrate the effectiveness of the proposed approach.

Online feature selection for multi-source streaming features

Multi-source streaming feature selection in an online manner has attracted considerable attention, from researchers because it can reduce the dimensionality of heterogeneous big data. However, traditional online algorithms such as Alpha-investing, Online Streaming Feature Selection (OSFS), Online Group Feature Selection (OGFS) and Scalable and Accurate OnLine Approach (SAOLA) consider only a single data source with fixed instances, and are not directly applicable to multi-source data. Multi-source Causal Feature Selection (MCFS) can search for an invariant set in multiple interventional datasets. However, fixed feature spaces are restrained, and exactly these same features are required among multi-source data. To overcome these limitations, we propose a novel method known as Multi-source Streaming Feature Selection (MSFS) to tackle the feature selection problem for multi-source streaming features. The MSFS algorithm addresses a new feature from a random source in three phases: relevance, intra-source redundancy, and inter-source redundancy analyses. That is, MSFS attempts to mine the potential relationships among different data sources rather than only independently consider each data source. In particular, each new feature is analyzed online using the overlapping instances from all data sources, and the Markov blanket (MB) of the target variable is dynamically adjusted. To evaluate the performance of the MSFS algorithm, we compare it with that of the abovementioned algorithms on 14 datasets and two real-world scenarios. The results demonstrate that MSFS outperforms the existing algorithms in classification accuracy and number of selected features.

Landslide susceptibility mapping: an integrated approach using knowledge-based numerical rating scheme, remote sensing, and multiple overlay analysis

Landslides are natural calamities that happen all over the world. Every year, landslides claim the lives of thousands of people and cause significant damage worldwide. The rapid and unpredictable expansion of many cities has a significant impact on the physical environment, with rapidly urbanizing regions of uneven topography where environmental circumstances make the construction of structures and infrastructure impossible or susceptible to instability. Our work proposes the integration of remote sensing, a knowledge-based numerical rating scheme, and multiple overlay analysis methodologies for landslide susceptibility mapping. We evaluate binary overlay, weighted overlay, and fuzzy overlay to map landslide vulnerability in the Ambegaon taluka of Pune district, Maharashtra. The input information used for the prediction and valuation of the landslide susceptibility map (LSM) includes an inventory of 39 active landslide points and 10 potential causality features of landslides: rainfall, slope, aspect, curvature, elevation, Euclidean distance to streamline, normalized difference vegetation index, topographic wetness index, Euclidean distance to roadline, and lithology. The area of consideration was categorized into five main susceptibility groups based on the computed landslide susceptibility scale, ranging from extremely low to severe. The LSM created using the fuzzy gamma operator (k = 0.95) has an overall forecast and prediction accuracy of 89.74%. The optimum value of the gamma operator (k) for this research is 0.95. The LSM was verified by correlating the frequency of landslides on different hazard classes. Implementation of this LSM at the regional level offers a foundation for expanding the approach to other geographies.

Causal Feature Selection with Missing Data

Causal feature selection aims at learning the Markov blanket (MB) of a class variable for feature selection. The MB of a class variable implies the local causal structure among the class variable and its MB and all other features are probabilistically independent of the class variable conditioning on its MB, this enables causal feature selection to identify potential causal features for feature selection for building robust and physically meaningful prediction models. Missing data, ubiquitous in many real-world applications, remain an open research problem in causal feature selection due to its technical complexity. In this article, we discuss a novel multiple imputation MB (MimMB) framework for causal feature selection with missing data. MimMB integrates Data Imputation with MB Learning in a unified framework to enable the two key components to engage with each other. MB Learning enables Data Imputation in a potentially causal feature space for achieving accurate data imputation, while accurate Data Imputation helps MB Learning identify a reliable MB of the class variable in turn. Then, we further design an enhanced kNN estimator for imputing missing values and instantiate the MimMB. In our comprehensively experimental evaluation, our new approach can effectively learn the MB of a given variable in a Bayesian network and outperforms other rival algorithms using synthetic and real-world datasets.

Error-aware Markov blanket learning for causal feature selection

Causal feature selection has attracted much attention in recent years, since it has better robustness than the traditional feature selection. Existing causal feature selection algorithms aim to identify a Markov blanket (MB) of the class variable. The MB of the class variable implies potential local causal relations around the class variable and has been proven to be the optimal feature subset for feature selection. Since almost all existing causal feature selection methods employ conditional independence (CI) tests to learn MBs, in practical settings, existing causal feature selection algorithms encounter the problem of CI test errors, which seriously deteriorates the performance of those existing methods. To solve this issue, in this paper, we propose an Error-Aware Markov Blanket learning (EAMB) algorithm with two novel subroutines to tackle the CI test error problem. Specifically, EAMB first identifies the MB of the class variable using one subroutine, and then utilizes the other subroutine to selectively recover the missed true MB features from the discarded features. The extensive experiments on 13 real-world datasets validate the effectiveness of EAMB against fourteen state-of-the-art causal feature selection algorithms and four well-established traditional feature selection methods.

A Light Causal Feature Selection Approach to High-Dimensional Data

Causal feature selection has received increasing attention in recent years. However, the state-of-the-art causal feature selection algorithms use the conditional independence tests, which require enumerating conditioning sets, leading to an exponential increase in computational complexity along with an increase in feature space. To address this problem, in this paper, we theoretically analyze the unique performance of causal features in mutual information, and propose a novel <underline xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">C</u> ausal <underline xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">F</u> eature <underline xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">S</u> election algorithm using <underline xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">M</u> utual <underline xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">I</u> nformation, called CFS-MI. Specifically, CFS-MI separately instantiates the pairwise comparison of mutual information in two stages to reduce computational complexity, and thus improves the efficiency on high-dimensional data. Extensive experiments on 5 benchmark Bayesian networks and 16 real-world datasets validate that CFS-MI has comparable accuracy compared to 7 state-of-the-art causal feature selection algorithms, while presenting more superior computational efficiency.

Suppressing reverberation in cochlear implant stimulus patterns using time-frequency masks based on phoneme groups.

Cochlear implant (CI) users experience considerable difficulty in understanding speech in reverberant listening environments. This issue is commonly addressed with time-frequency masking, where a time-frequency decomposed reverberant signal is multiplied by a matrix of gain values to suppress reverberation. However, mask estimation is challenging in reverberant environments due to the large spectro-temporal variations in the speech signal. To overcome this variability, we previously developed a phoneme-based algorithm that selects a different mask estimation model based on the underlying phoneme. In the ideal case where knowledge of the phoneme was assumed, the phoneme-based approach provided larger benefits than a phoneme-independent approach when tested in normal-hearing listeners using an acoustic model of CI processing. The current work investigates the phoneme-based mask estimation algorithm in the real-time feasible case where the prediction from a phoneme classifier is used to select the phoneme-specific mask. To further ensure real-time feasibility, both the phoneme classifier and mask estimation algorithm use causal features extracted from within the CI processing framework. We conducted experiments in normal-hearing listeners using an acoustic model of CI processing, and the results showed that the phoneme-specific algorithm benefitted the majority of subjects.

Causal Feature Learning

Causal Feature Learning

Correlation-based and feature-driven mutation signature analyses to identify genetic features associated with DNA mutagenic processes in cancer genomes.

Mutation signatures represent unique sequence footprints of somatic mutations resulting from specific DNA mutagenic and repair processes. However, their causal associations and the potential utility for genome research remain largely unknown. In this study, we performed PanCancer-scale correlative analyses to identify the genomic features associated with tumor mutation burdens (TMB) and individual mutation signatures. We observed that TMB was correlated with tumor purity, ploidy, and the level of aneuploidy, as well as with the expression of cell proliferation-related genes representing genomic covariates in evaluating TMB. Correlative analyses of mutation signature levels with genes belonging to specific DNA damage-repair processes revealed that deficiencies of NHEJ1 and ALKBH3 may contribute to mutations in the settings of APOBEC cytidine deaminase activation and DNA mismatch repair deficiency, respectively. We further employed a strategy to identify feature-driven, de novo mutation signatures and demonstrated that mutation signatures can be reconstructed using known causal features. Using the strategy, we further identified tumor hypoxia-related mutation signatures similar to the APOBEC-related mutation signatures, suggesting that APOBEC activity mediates hypoxia-related mutational consequences in cancer genomes. Our study advances the mechanistic insights into the TMB and signature-based DNA mutagenic and repair processes in cancer genomes. We also propose that feature-driven mutation signature analysis can further extend the categories of cancer-relevant mutation signatures and their causal relationships.

Implementation as Resemblance

This article advertises a new account of computational implementation. According to the resemblance account, implementation is a matter of resembling a computational architecture. The resemblance account departs from previous theories by denying that computational architectures are exhausted by their formal, mathematical features. Instead, they are taken to be permeated with causality, spatiotemporality, and other nonmathematical features. I argue that this approach comports well with computer scientific practice and offers a novel response to so-called triviality arguments.

Explaining and predicting human behavior and social dynamics in simulated virtual worlds: reproducibility, generalizability, and robustness of causal discovery methods

Ground Truth program was designed to evaluate social science modeling approaches using simulation test beds with ground truth intentionally and systematically embedded to understand and model complex Human Domain systems and their dynamics Lazer et al. (Science 369:1060–1062, 2020). Our multidisciplinary team of data scientists, statisticians, experts in Artificial Intelligence (AI) and visual analytics had a unique role on the program to investigate accuracy, reproducibility, generalizability, and robustness of the state-of-the-art (SOTA) causal structure learning approaches applied to fully observed and sampled simulated data across virtual worlds. In addition, we analyzed the feasibility of using machine learning models to predict future social behavior with and without causal knowledge explicitly embedded. In this paper, we first present our causal modeling approach to discover the causal structure of four virtual worlds produced by the simulation teams—Urban Life, Financial Governance, Disaster and Geopolitical Conflict. Our approach adapts the state-of-the-art causal discovery (including ensemble models), machine learning, data analytics, and visualization techniques to allow a human-machine team to reverse-engineer the true causal relations from sampled and fully observed data. We next present our reproducibility analysis of two research methods team’s performance using a range of causal discovery models applied to both sampled and fully observed data, and analyze their effectiveness and limitations. We further investigate the generalizability and robustness to sampling of the SOTA causal discovery approaches on additional simulated datasets with known ground truth. Our results reveal the limitations of existing causal modeling approaches when applied to large-scale, noisy, high-dimensional data with unobserved variables and unknown relationships between them. We show that the SOTA causal models explored in our experiments are not designed to take advantage from vasts amounts of data and have difficulty recovering ground truth when latent confounders are present; they do not generalize well across simulation scenarios and are not robust to sampling; they are vulnerable to data and modeling assumptions, and therefore, the results are hard to reproduce. Finally, when we outline lessons learned and provide recommendations to improve models for causal discovery and prediction of human social behavior from observational data, we highlight the importance of learning data to knowledge representations or transformations to improve causal discovery and describe the benefit of causal feature selection for predictive and prescriptive modeling.

Accelerating Causal Inference and Feature Selection Methods through G-Test Computation Reuse.

This article presents a novel and remarkably efficient method of computing the statistical G-test made possible by exploiting a connection with the fundamental elements of information theory: by writing the G statistic as a sum of joint entropy terms, its computation is decomposed into easily reusable partial results with no change in the resulting value. This method greatly improves the efficiency of applications that perform a series of G-tests on permutations of the same features, such as feature selection and causal inference applications because this decomposition allows for an intensive reuse of these partial results. The efficiency of this method is demonstrated by implementing it as part of an experiment involving IPC–MB, an efficient Markov blanket discovery algorithm, applicable both as a feature selection algorithm and as a causal inference method. The results show outstanding efficiency gains for IPC–MB when the G-test is computed with the proposed method, compared to the unoptimized G-test, but also when compared to IPC–MB++, a variant of IPC–MB which is enhanced with an AD–tree, both static and dynamic. Even if this proposed method of computing the G-test is presented here in the context of IPC–MB, it is in fact bound neither to IPC–MB in particular, nor to feature selection or causal inference applications in general, because this method targets the information-theoretic concept that underlies the G-test, namely conditional mutual information. This aspect grants it wide applicability in data sciences.

Rehabilitation Tracking of Athletes Post Anterior Cruciate Ligament Reconstruction (ACL-R) Surgery Through Causal Analysis of Gait Data & Computational Modeling.

Early identification of motion disparities in Anterior Cruciate Ligament reconstructed (ACL-R) athletes may better post-operative decision making when returning athletes to sport. Existing return to play assessments consist of assessments of muscle strength, functional tasks, patient-reported outcomes, and 3D coordinate tracking. However, these methods primarily depend on the medical provider's intuition to release them to participate in an unrestricted activity after ACL-R that may cause reinjury or long-term impacts. This study proposes a wearable sensor-based system that helps track athlete rehabilitation progress and return to sport decision making. For this, we capture gait data from 89 ACL-R athletes during their walking and jogging trials. The raw gyroscope data collected from this system is used to extract causal features based on Nolte's phase slope index. Features extracted from this study are used to develop computational models that classify ACL-R athletes based on their reconstructed knee during two visits (3-6 months & 9 months) post ACL-R surgery. The classifier's performance degradation in detecting ACL-R athletes injured knee during multiple visits supports athletic trainers and physicians' decision-making process to confirm an athlete's safe return to sport.Clinical Relevance- This study develops computational models based on causal analysis of gait data to support athletic trainers and medical practitioners' decision to return athletes to sport post ACL-R surgery.

Metaphorical models and the translator’ s approach to scientific texts

Indeed, the utility of theory-constitutive metaphor seems to lie largely in the fact that they provide a way to introduce terminology for features of the world whose existence seems probable, but many of whose fundamental properties have yet to be discovered. Theory constitutive metaphors, in other words, represent one strategy for the accommodation of language to as yet undiscovered causal features of the world. (Boyd 1979:364)In this article we investigate the mechanisms behind the urge for better and new understanding. Our hypothesis is that these are generally related to and inspired by metaphorical reasoning. The existing cognitive model of a source domain of understanding is used to structure and categorise a target domain. The metaphorical model is an underlying schema which is not fully expressed propositionally and lexically. The traces and results of metapho¬rical thinking are in metaphorical lexicalisations. We challenge the princi¬ple of traditional Vienna school terminology theory which claims that because unambiguous communication is the ideal in scientific communica¬tion, it is preferable to replace a metaphorical term by its literal equivalent. We show how in the life sciences the gestalt-like metaphorical model which was at the basis of better understanding and new discovery continues to have an important role in didactic as well as in scientific texts which treat the same or related subjects. The discipline of terminology theory could work out guidelines for the description of metaphorical models starting from the metaphorical lexicalisations which are the traces which metaphorical think¬ing leaves in a language. Translators who are aware of the impact of metaphorical reasoning on lexicalisation in scientific language will develop the ability to distinguish between metaphorical models which are language and culture independent and those which are not.