Baseline functional connectivity predicts who will benefit from neuromodulation: evidence from primary progressive aphasia.

Background: Identifying the characteristics of individuals who demonstrate response to an intervention allows us to predict who is most likely to benefit from certain interventions. Prediction is challenging in rare and heterogeneous diseases, such as primary progressive aphasia (PPA), that have varying clinical manifestations. We aimed to determine the characteristics of those who will benefit most from transcranial direct current stimulation (tDCS) of the left inferior frontal gyrus (IFG) using a novel heterogeneity and group identification analysis. Methods: We compared the predictive ability of demographic and clinical patient characteristics (e.g., PPA variant and disease progression, baseline language performance) vs. functional connectivity alone (from resting-state fMRI) in the same cohort. Results: Functional connectivity alone had the highest predictive value for outcomes, explaining 62% and 75% of tDCS effect of variance in generalization (semantic fluency) and in the trained outcome of the clinical trial (written naming), contrasted with <15% predicted by clinical characteristics, including baseline language performance. Patients with higher baseline functional connectivity between the left IFG (opercularis and triangularis), and between the middle temporal pole and posterior superior temporal gyrus, were most likely to benefit from tDCS. Conclusions: We show the importance of a baseline 7-minute functional connectivity scan in predicting tDCS outcomes, and point towards a precision medicine approach in neuromodulation studies. The study has important implications for clinical trials and practice, providing a statistical method that addresses heterogeneity in patient populations and allowing accurate prediction and enrollment of those who will most likely benefit from specific interventions.

in silico transcriptome dissection of neocortical excitatory neurogenesis via joint matrix decomposition and transfer learning.

The rising quality and amount of multi-omic data across biomedical science demands that we build innovative solutions to harness their collective discovery potential. From publicly available repositories, we have assembled and curated a compendium of gene-level transcriptomic data focused on mammalian excitatory neurogenesis in the neocortex. This collection is open for exploration by both computational and cell biologists at nemoanalytics.org, and this report forms a demonstration of its utility. Applying our novel structured joint decomposition approach to mouse, macaque and human data from the collection, we define transcriptome dynamics that are conserved across mammalian excitatory neurogenesis and which map onto the genetics of human brain structure and disease. Leveraging additional data within NeMO Analytics via projection methods, we chart the dynamics of these fundamental molecular elements of neurogenesis across developmental time and space and into postnatal life. Reversing the direction of our investigation, we use transcriptomic data from laminar-specific dissection of adult human neocortex to define molecular signatures specific to excitatory neuronal cell types resident in individual layers of the mature neocortex, and trace their emergence across development. We show that while many lineage defining transcription factors are most highly expressed at early fetal ages, the laminar neuronal identities which they drive take years to decades to reach full maturity. Finally, we interrogated data from stem-cell derived cerebral organoid systems demonstrating that many fundamental elements of in vivo development are recapitulated with high-fidelity in vitro, while specific transcriptomic programs in neuronal maturation are absent. We propose these analyses as specific applications of the general approach of combining joint decomposition with large curated collections of analysis-ready multi-omics data matrices focused on particular cell and disease contexts. Importantly, these open environments are accessible to, and must be fueled with emerging data by, cell biologists with and without coding expertise.

Mapping Colorectal Distribution of Cell-free and Cell-associated HIV Surrogates Following Simulated Anal Intercourse to Aid Rectal Microbicide Development.

BACKGROUND: Anal sex remains the greatest HIV transmission risk for men who have sex with men and carries substantial population attributable risk among women. Despite a growing array of HIV pre-exposure prophylaxis (PrEP) options, rectal microbicides remain desirable as on demand, non-systemic PrEP. Rectal microbicide product development for PrEP requires understanding the spatiotemporal distribution of HIV infectious elements in the rectosigmoid to optimize formulation development. SETTING: Outpatient setting with healthy research participants. METHODS: Six healthy men underwent simulated receptive anal sex with an artificial phallus fitted with a triple lumen catheter in the urethral position. To simulate ejaculation of HIV-infected semen, autologous seminal plasma laden with autologous blood lymphocytes from apheresis labeled with 111Indium-oxine (cell-associated) and 99mTechnetium-sulfur colloid (cell-free) as HIV surrogates were injected into the rectal lumen through the phallic urethra. Spatiotemporal distribution of each radioisotope was assessed using SPECT/CT over eight hours. Analysis of radiolabel distribution used a flexible principal curve algorithm to quantitatively estimate rectal lumen distribution. RESULTS: Cell-free and cell-associated HIV surrogates distributed to a maximal distance of 15 and 16 cm, respectively, from the anorectal junction (∼19 and ∼20 cm from the anal verge), with a maximal signal intensity located 6 and 7 cm, respectively. There were no significant differences in any distribution parameters between cell-free and cell-associated HIV surrogate. CONCLUSIONS: Cell-free and cell-associated HIV surrogate distribution in the rectosigmoid can be quantified with spatiotemporal pharmacokinetic methods. These results describe the ideal luminal target distribution to guide rectal microbicide development.

Spatial domain detection using contrastive self-supervised learning for spatial multi-omics technologies.

Recent advances in spatially-resolved single-omics and multi-omics technologies have led to the emergence of computational tools to detect or predict spatial domains. Additionally, histological images and immunofluorescence (IF) staining of proteins and cell types provide multiple perspectives and a more complete understanding of tissue architecture. Here, we introduce Proust, a scalable tool to predict discrete domains using spatial multi-omics data by combining the low-dimensional representation of biological profiles based on graph-based contrastive self-supervised learning. Our scalable method integrates multiple data modalities, such as RNA, protein, and H&E images, and predicts spatial domains within tissue samples. Through the integration of multiple modalities, Proust consistently demonstrates enhanced accuracy in detecting spatial domains, as evidenced across various benchmark datasets and technological platforms.

A machine learning-based choledocholithiasis prediction tool to improve ERCP decision making: a proof-of-concept study.

BACKGROUND: Previous studies demonstrated limited accuracy of existing guidelines for predicting choledocholithiasis, leading to overutilization of endoscopic retrograde cholangiopancreatography (ERCP). More accurate stratification may improve patient selection for ERCP and allow use of lower-risk modalities. METHODS: A machine learning model was developed using patient information from two published cohort studies that evaluated performance of guidelines in predicting choledocholithiasis. Prediction models were developed using the gradient boosting model (GBM) machine learning method. GBM performance was evaluated using 10-fold cross-validation and area under the receiver operating characteristic curve (AUC). Important predictors of choledocholithiasis were identified based on relative importance in the GBM. RESULTS: 1378 patients (mean age 43.3 years; 61.2% female) were included in the GBM and 59.4% had choledocholithiasis. Eight variables were identified as predictors of choledocholithiasis. The GBM had accuracy of 71.5% (SD 2.5%) (AUC 0.79 [SD 0.06]) and performed better than the 2019 American Society for Gastrointestinal Endoscopy (ASGE) guidelines (accuracy 62.4% [SD 2.6%]; AUC 0.63 [SD 0.03]) and European Society of Gastrointestinal Endoscopy (ESGE) guidelines (accuracy 62.8% [SD 2.6%]; AUC 0.67 [SD 0.02]). The GBM correctly categorized 22% of patients directed to unnecessary ERCP by ASGE guidelines, and appropriately recommended as the next management step 48% of ERCPs incorrectly rejected by ESGE guidelines. CONCLUSIONS: A machine learning-based tool was created, providing real-time, personalized, objective probability of choledocholithiasis and ERCP recommendations. This more accurately directed ERCP use than existing ASGE and ESGE guidelines, and has the potential to reduce morbidity associated with ERCP or missed choledocholithiasis.

Regression models for partially localized fMRI connectivity analyses.

Background: Brain functional connectivity analysis of resting-state functional magnetic resonance imaging (fMRI) data is typically performed in a standardized template space assuming consistency of connections across subjects. Analysis methods can come in the form of one-edge-at-a-time analyses or dimension reduction/decomposition methods. Common to these approaches is an assumption that brain regions are functionally aligned across subjects; however, it is known that this functional alignment assumption is often violated. Methods: In this paper, we use subject-level regression models to explain intra-subject variability in connectivity. Covariates can include factors such as geographic distance between two pairs of brain regions, whether the two regions are symmetrically opposite (homotopic), and whether the two regions are members of the same functional network. Additionally, a covariate for each brain region can be included, to account for the possibility that some regions have consistently higher or lower connectivity. This style of analysis allows us to characterize the fraction of variation explained by each type of covariate. Additionally, comparisons across subjects can then be made using the fitted connectivity regression models, offering a more parsimonious alternative to edge-at-a-time approaches. Results: We apply our approach to Human Connectome Project data on 268 regions of interest (ROIs), grouped into eight functional networks. We find that a high proportion of variation is explained by region covariates and network membership covariates, while geographic distance and homotopy have high relative importance after adjusting for the number of predictors. We also find that the degree of data repeatability using our connectivity regression model-which uses only partial location information about pairs of ROI's-is comparably as high as the repeatability obtained using full location information. Discussion: While our analysis uses data that have been transformed into a common template-space, we also envision the method being useful in multi-atlas registration settings, where subject data remains in its own geometry and templates are warped instead. These results suggest the tantalizing possibility that fMRI connectivity analysis can be performed in subject-space, using less aggressive registration, such as simple affine transformations, multi-atlas subject-space registration, or perhaps even no registration whatsoever.

Longitudinal trajectories of glycemic control among U.S. Adults with newly diagnosed diabetes.

AIMS: To identify longitudinal trajectories of glycemic control among adults with newly diagnosed diabetes, overall and by diabetes type. METHODS: We analyzed claims data from OptumLabs® Data Warehouse for 119,952 adults newly diagnosed diabetes between 2005 and 2018. We applied a novel Mixed Effects Machine Learning model to identify longitudinal trajectories of hemoglobin A1c (HbA1c) over 3 years of follow-up and used multinomial regression to characterize factors associated with each trajectory. RESULTS: The study population was comprised of 119,952 adults with newly diagnosed diabetes, including 696 (0.58%) with type 1 diabetes. Among patients with type 1 diabetes, 52.6% were diagnosed at very high HbA1c, partially improved, but never achieved control; 32.5% were diagnosed at low HbA1c and deteriorated over time; and 14.9% had stable low HbA1c. Among patients with type 2 diabetes, 67.7% had stable low HbA1c, 14.4% were diagnosed at very high HbA1c, partially improved, but never achieved control; 10.0% were diagnosed at moderately high HbA1c and deteriorated over time; and 4.9% were diagnosed at moderately high HbA1c and improved over time. CONCLUSIONS: Claims data identified distinct longitudinal trajectories of HbA1c after diabetes diagnosis, which can be used to anticipate challenges and individualize care plans to improve glycemic control.

A Multi-Trait Multi-Method Examination of Psychometric Instrument Performance in Autism Spectrum Disorder.

Anecdotal evidence has suggested that rater-based measures (e.g., parent report) may have strong across-trait/within-individual covariance that detracts from trait-specific measurement precision; rater measurement-related bias may help explain poor correlation within Autism Spectrum Disorder (ASD) samples between rater-based and performance-based measures of the same trait. We used a multi-trait, multi-method approach to examine method-associated bias within an ASD sample (n = 83). We examined performance/rater-instrument pairs for attention, inhibition, working memory, motor coordination, and core ASD features. Rater-based scores showed an overall greater methodology bias (57% of variance in score explained by method), while performance-based scores showed a weaker methodology bias (22%). The degree of inter-individual variance explained by method alone substantiates an anecdotal concern associated with the use of rater measures in ASD.

Timing matters: The contribution of running during different periods of the light/dark cycle to susceptibility to activity-based anorexia in rats.

Individuals with anorexia nervosa (AN) exhibit dangerous weight loss due to restricted eating and hyperactivity. Those with AN are predominantly women and most cases have an age of onset during adolescence. Activity-based anorexia (ABA) is a rodent behavioral paradigm that recapitulates many of the features of AN including restricted food intake and hyperactivity, resulting in precipitous weight loss. In addition, there is enhanced sensitivity to the paradigm during adolescence. In ABA, animals are given time-restricted access to food and unlimited access to a running wheel. Under these conditions, most animals increase their running and decrease their food intake resulting in precipitous weight loss until they either die or researchers discontinue the paradigm. Some animals learn to balance their food intake and energy expenditure and are able to stabilize and eventually reverse their weight loss. For these studies, adolescent (postnatal day 33-42), female Sprague Dawley (n = 68) rats were placed under ABA conditions (unlimited access to a running wheel and 1.5 hrs access to food) until they either reached 25% body weight loss or for 7 days. 70.6% of subjects reached 25% body weight loss before 7 days and were designated susceptible to ABA while 29.4% animals were resistant to the paradigm and did not achieve the weight loss criterion. We used discrete time survival analysis to investigate the contribution of food intake and running behavior during distinct time periods both prior to and during ABA to the likelihood of reaching the weight loss criterion and dropping out of ABA. Our analyses revealed risk factors, including total running and dark cycle running, that increased the likelihood of dropping out of the paradigm, as well as protective factors, including age at the start of ABA, the percent of total running exhibited as food anticipatory activity (FAA), and food intake, that reduced the likelihood of dropping out. These measures had predictive value whether taken before or during exposure to ABA conditions. Our findings suggest that certain running and food intake behaviors may be indicative of a phenotype that predisposes animals to susceptibility to ABA. They also provide evidence that running during distinct time periods may reflect functioning of distinct neural circuitry and differentially influence susceptibility and resistance to the paradigm.

Specificity in Generalization Effects of Transcranial Direct Current Stimulation Over the Left Inferior Frontal Gyrus in Primary Progressive Aphasia.

OBJECTIVES: Generalization (or near-transfer) effects of an intervention to tasks not explicitly trained are the most desirable intervention outcomes. However, they are rarely reported and even more rarely explained. One hypothesis for generalization effects is that the tasks improved share the same brain function/computation with the intervention task. We tested this hypothesis in this study of transcranial direct current stimulation (tDCS) over the left inferior frontal gyrus (IFG) that is claimed to be involved in selective semantic retrieval of information from the temporal lobes. MATERIALS AND METHODS: In this study, we examined whether tDCS over the left IFG in a group of patients with primary progressive aphasia (PPA), paired with a lexical/semantic retrieval intervention (oral and written naming), may specifically improve semantic fluency, a nontrained near-transfer task that relies on selective semantic retrieval, in patients with PPA. RESULTS: Semantic fluency improved significantly more in the active tDCS than in the sham tDCS condition immediately after and two weeks after treatment. This improvement was marginally significant two months after treatment. We also found that the active tDCS effect was specific to tasks that require this IFG computation (selective semantic retrieval) but not to other tasks that may require different computations of the frontal lobes. CONCLUSIONS: We provided interventional evidence that the left IFG is critical for selective semantic retrieval, and tDCS over the left IFG may have a near-transfer effect on tasks that depend on the same computation, even if they are not specifically trained. CLINICAL TRIAL REGISTRATION: The Clinicaltrials.gov registration number for the study is NCT02606422.

Predicting chronic postsurgical pain: current evidence and a novel program to develop predictive biomarker signatures.

ABSTRACT: Chronic pain affects more than 50 million Americans. Treatments remain inadequate, in large part, because the pathophysiological mechanisms underlying the development of chronic pain remain poorly understood. Pain biomarkers could potentially identify and measure biological pathways and phenotypical expressions that are altered by pain, provide insight into biological treatment targets, and help identify at-risk patients who might benefit from early intervention. Biomarkers are used to diagnose, track, and treat other diseases, but no validated clinical biomarkers exist yet for chronic pain. To address this problem, the National Institutes of Health Common Fund launched the Acute to Chronic Pain Signatures (A2CPS) program to evaluate candidate biomarkers, develop them into biosignatures, and discover novel biomarkers for chronification of pain after surgery. This article discusses candidate biomarkers identified by A2CPS for evaluation, including genomic, proteomic, metabolomic, lipidomic, neuroimaging, psychophysical, psychological, and behavioral measures. Acute to Chronic Pain Signatures will provide the most comprehensive investigation of biomarkers for the transition to chronic postsurgical pain undertaken to date. Data and analytic resources generatedby A2CPS will be shared with the scientific community in hopes that other investigators will extract valuable insights beyond A2CPS's initial findings. This article will review the identified biomarkers and rationale for including them, the current state of the science on biomarkers of the transition from acute to chronic pain, gaps in the literature, and how A2CPS will address these gaps.

Template independent component analysis with spatial priors for accurate subject-level brain network estimation and inference.

Independent component analysis is commonly applied to functional magnetic resonance imaging (fMRI) data to extract independent components (ICs) representing functional brain networks. While ICA produces reliable group-level estimates, single-subject ICA often produces noisy results. Template ICA is a hierarchical ICA model using empirical population priors to produce more reliable subject-level estimates. However, this and other hierarchical ICA models assume unrealistically that subject effects are spatially independent. Here, we propose spatial template ICA (stICA), which incorporates spatial priors into the template ICA framework for greater estimation efficiency. Additionally, the joint posterior distribution can be used to identify brain regions engaged in each network using an excursions set approach. By leveraging spatial dependencies and avoiding massive multiple comparisons, stICA has high power to detect true effects. We derive an efficient expectation-maximization algorithm to obtain maximum likelihood estimates of the model parameters and posterior moments of the latent fields. Based on analysis of simulated data and fMRI data from the Human Connectome Project, we find that stICA produces estimates that are more accurate and reliable than benchmark approaches, and identifies larger and more reliable areas of engagement. The algorithm is computationally tractable, achieving convergence within 12 hours for whole-cortex fMRI analysis.

Regression models for partially localized fMRI connectivity analyses.

Brain functional connectivity analysis of resting-state functional magnetic resonance imaging (fMRI) data is typically performed in a standardized template space assuming consistency of connections across subjects. This can come in the form of one-edge-at-a-time analyses or dimension reduction/decomposition methods. Common to these approaches is the assumption of complete localization (or spatial alignment) of brain regions across subjects. Alternative approaches completely eschew localization assumptions by treating connections as statistically exchangeable (for example, using the density of connectivity between nodes). Yet other approaches, such as hyperalignment, attempt to align subjects on function as well as structure, thereby achieving a different sort of template-based localization. In this paper, we propose the use of simple regression models to characterize connectivity. To that end, we build regression models on subject-level Fisher transformed regional connection matrices using geographic distance, homotopic distance, network labels, and region indicators as covariates to explain variation in connections. While we perform our analysis in template-space in this paper, we envision the method being useful in multi-atlas registration settings, where subject data remains in its own geometry and templates are warped instead. A byproduct of this style of analysis is the ability to characterize the fraction of variation in subject-level connections explained by each type of covariate. Using Human Connectome Project data, we found that network labels and regional characteristics contribute far more than geographic or homotopic relationships (considered non-parametrically). In addition, visual regions had the highest explanatory power (i.e., largest regression coefficients). We also considered subject repeatability and found that the degree of repeatability seen in fully localized models is largely recovered using our proposed subject-level regression models. Further, even fully exchangeable models retain a sizeable amount of repeatability information, despite discarding all localization information. These results suggest the tantalizing possibility that fMRI connectivity analysis can be performed in subject-space, using less aggressive registration, such as simple affine transformations, multi-atlas subject-space registration, or perhaps even no registration whatsoever.

Autism and Hierarchical Models of Intelligence.

BACKGROUND: The Wechsler Intelligence Scale for Children (WISC) employs a hierarchical model of general intelligence in which index scores separate out different clinically-relevant aspects of intelligence; the test is designed such that index scores are statistically independent from one another within the normative sample. Whether or not the existing index scores meet the desired psychometric property of being statistically independent within autistic samples is unknown. METHOD: We conducted a factor analysis on WISC fifth edition (WISC-V) (N = 83) and WISC fourth edition (WISC-IV) (N = 131) subtest data in children with autism. We compared the data-driven exploratory factor analysis with the manual-derived index scores, including in a typically developing (TD) WISC-IV cohort (N = 209). RESULTS: The WISC-IV TD cohort showed the expected 1:1 relationship between empirically derived factors and manual-derived index scores. We observed less unique correlations between our data-driven factors and manualized IQ index scores in both ASD samples (WISC-IV and WISC-V). In particular, in both WISC-IV and -V, working memory (WM) influenced index scores in autistic individuals that do not load on WM in the normative sample. CONCLUSIONS: WISC index scores do not show the desired statistical independence within autistic samples, as judged against an empirically-derived exploratory factor analysis. In particular, within the currently used WISC-V version, WM influences multiple index scores.

Corrigendum: Editorial: Explainable artificial intelligence models and methods in finance and healthcare.

[This corrects the article DOI: 10.3389/frai.2022.970246.].

Evaluating causal psychological models: A study of language theories of autism using a large sample.

We used a large convenience sample (n = 22,223) from the Simons Powering Autism Research (SPARK) dataset to evaluate causal, explanatory theories of core autism symptoms. In particular, the data-items collected supported the testing of theories that posited altered language abilities as cause of social withdrawal, as well as alternative theories that competed with these language theories. Our results using this large dataset converge with the evolution of the field in the decades since these theories were first proposed, namely supporting primary social withdrawal (in some cases of autism) as a cause of altered language development, rather than vice versa. To accomplish the above empiric goals, we used a highly theory-constrained approach, one which differs from current data-driven modeling trends but is coherent with a very recent resurgence in theory-driven psychology. In addition to careful explication and formalization of theoretical accounts, we propose three principles for future work of this type: specification, quantification, and integration. Specification refers to constraining models with pre-existing data, from both outside and within autism research, with more elaborate models and more veridical measures, and with longitudinal data collection. Quantification refers to using continuous measures of both psychological causes and effects, as well as weighted graphs. This approach avoids "universality and uniqueness" tests that hold that a single cognitive difference could be responsible for a heterogeneous and complex behavioral phenotype. Integration of multiple explanatory paths within a single model helps the field examine for multiple contributors to a single behavioral feature or to multiple behavioral features. It also allows integration of explanatory theories across multiple current-day diagnoses and as well as typical development.

Rethinking recovery in adolescent concussions: Network-level functional connectivity alterations associated with motor deficits.

Adolescents who are clinically recovered from concussion continue to show subtle motor impairment on neurophysiological and behavioral measures. However, there is limited information on brain-behavior relationships of persistent motor impairment following clinical recovery from concussion. We examined the relationship between subtle motor performance and functional connectivity of the brain in adolescents with a history of concussion, status post-symptom resolution, and subjective return to baseline. Participants included 27 adolescents who were clinically recovered from concussion and 29 never-concussed, typically developing controls (10-17 years); all participants were examined using the Physical and Neurologic Examination of Subtle Signs (PANESS). Functional connectivity between the default mode network (DMN) or dorsal attention network (DAN) and regions of interest within the motor network was assessed using resting-state functional magnetic resonance imaging (rsfMRI). Compared to controls, adolescents clinically recovered from concussion showed greater subtle motor deficits as evaluated by the PANESS and increased connectivity between the DMN and left lateral premotor cortex. DMN to left lateral premotor cortex connectivity was significantly correlated with the total PANESS score, with more atypical connectivity associated with more motor abnormalities. This suggests that altered functional connectivity of the brain may underlie subtle motor deficits in adolescents who have clinically recovered from concussion. More investigation is required to understand the persistence and longer-term clinical relevance of altered functional connectivity and associated subtle motor deficits to inform whether functional connectivity may serve as an important biomarker related to longer-term outcomes after clinical recovery from concussion.

First Organoid Intelligence (OI) workshop to form an OI community.

The brain is arguably the most powerful computation system known. It is extremely efficient in processing large amounts of information and can discern signals from noise, adapt, and filter faulty information all while running on only 20 watts of power. The human brain's processing efficiency, progressive learning, and plasticity are unmatched by any computer system. Recent advances in stem cell technology have elevated the field of cell culture to higher levels of complexity, such as the development of three-dimensional (3D) brain organoids that recapitulate human brain functionality better than traditional monolayer cell systems. Organoid Intelligence (OI) aims to harness the innate biological capabilities of brain organoids for biocomputing and synthetic intelligence by interfacing them with computer technology. With the latest strides in stem cell technology, bioengineering, and machine learning, we can explore the ability of brain organoids to compute, and store given information (input), execute a task (output), and study how this affects the structural and functional connections in the organoids themselves. Furthermore, understanding how learning generates and changes patterns of connectivity in organoids can shed light on the early stages of cognition in the human brain. Investigating and understanding these concepts is an enormous, multidisciplinary endeavor that necessitates the engagement of both the scientific community and the public. Thus, on Feb 22-24 of 2022, the Johns Hopkins University held the first Organoid Intelligence Workshop to form an OI Community and to lay out the groundwork for the establishment of OI as a new scientific discipline. The potential of OI to revolutionize computing, neurological research, and drug development was discussed, along with a vision and roadmap for its development over the coming decade.

A machine learning based approach towards high-dimensional mediation analysis.

Mediation analysis is used to investigate the role of intermediate variables (mediators) that lie in the path between an exposure and an outcome variable. While significant research has focused on developing methods for assessing the influence of mediators on the exposure-outcome relationship, current approaches do not easily extend to settings where the mediator is high-dimensional. These situations are becoming increasingly common with the rapid increase of new applications measuring massive numbers of variables, including brain imaging, genomics, and metabolomics. In this work, we introduce a novel machine learning based method for identifying high dimensional mediators. The proposed algorithm iterates between using a machine learning model to map the high-dimensional mediators onto a lower-dimensional space, and using the predicted values as input in a standard three-variable mediation model. Hence, the machine learning model is trained to maximize the likelihood of the mediation model. Importantly, the proposed algorithm is agnostic to the machine learning model that is used, providing significant flexibility in the types of situations where it can be used. We illustrate the proposed methodology using data from two functional Magnetic Resonance Imaging (fMRI) studies. First, using data from a task-based fMRI study of thermal pain, we combine the proposed algorithm with a deep learning model to detect distributed, network-level brain patterns mediating the relationship between stimulus intensity (temperature) and reported pain at the single trial level. Second, using resting-state fMRI data from the Human Connectome Project, we combine the proposed algorithm with a connectome-based predictive modeling approach to determine brain functional connectivity measures that mediate the relationship between fluid intelligence and working memory accuracy. In both cases, our multivariate mediation model links exposure variables (thermal pain or fluid intelligence), high dimensional brain measures (single-trial brain activation maps or resting-state brain connectivity) and behavioral outcomes (pain report or working memory accuracy) into a single unified model. Using the proposed approach, we are able to identify brain-based measures that simultaneously encode the exposure variable and correlate with the behavioral outcome.

Differences in functional connectivity distribution after transcranial direct-current stimulation: A connectivity density point of view.

In this manuscript, we consider the problem of relating functional connectivity measurements viewed as statistical distributions to outcomes. We demonstrate the utility of using the distribution of connectivity on a study of resting-state functional magnetic resonance imaging association with an intervention. The method uses the estimated density of connectivity between nodes of interest as a functional covariate. Moreover, we demonstrate the utility of the procedure in an instance where connectivity is naturally considered an outcome by reversing the predictor/response relationship using case/control methodology. The method utilizes the density quantile, the density evaluated at empirical quantiles, instead of the empirical density directly. This improved the performance of the method by highlighting tail behavior, though we emphasize that by being flexible and non-parametric, the technique can detect effects related to the central portion of the density. To demonstrate the method in an application, we consider 47 primary progressive aphasia patients with various levels of language abilities. These patients were randomly assigned to two treatment arms, transcranial direct-current stimulation and language therapy versus sham (language therapy only), in a clinical trial. We use the method to analyze the effect of direct stimulation on functional connectivity. As such, we estimate the density of correlations among the regions of interest and study the difference in the density post-intervention between treatment arms. We discover that it is the tail of the density, rather than the mean or lower order moments of the distribution, that demonstrates a significant impact in the classification. The new approach has several benefits. Among them, it drastically reduces the number of multiple comparisons compared with edge-wise analysis. In addition, it allows for the investigation of the impact of functional connectivity on the outcomes where the connectivity is not geometrically localized.