Computational Methods for Predicting Autism Spectrum Disorder from Gene Expression Data

BackgroundIn recent years, human microbiome studies have received increasing attention as this field is considered a potential source for clinical applications. With the advancements in omics technologies and AI, research focused on the discovery for potential biomarkers in the human microbiome using machine learning tools has produced positive outcomes. Despite the promising results, several issues can still be found in these studies such as datasets with small number of samples, inconsistent results, lack of uniform processing and methodologies, and other additional factors lead to lack of reproducibility in biomedical research. In this work, we propose a methodology that combines the DADA2 pipeline for 16s rRNA sequences processing and the Recursive Ensemble Feature Selection (REFS) in multiple datasets to increase reproducibility and obtain robust and reliable results in biomedical research.ResultsThree experiments were performed analyzing microbiome data from patients/cases in Inflammatory Bowel Disease (IBD), Autism Spectrum Disorder (ASD), and Type 2 Diabetes (T2D). In each experiment, we found a biomarker signature in one dataset and applied to 2 other as further validation. The effectiveness of the proposed methodology was compared with other feature selection methods such as K-Best with F-score and random selection as a base line. The Area Under the Curve (AUC) was employed as a measure of diagnostic accuracy and used as a metric for comparing the results of the proposed methodology with other feature selection methods. Additionally, we use the Matthews Correlation Coefficient (MCC) as a metric to evaluate the performance of the methodology as well as for comparison with other feature selection methods.ConclusionsWe developed a methodology for reproducible biomarker discovery for 16s rRNA microbiome sequence analysis, addressing the issues related with data dimensionality, inconsistent results and validation across independent datasets. The findings from the three experiments, across 9 different datasets, show that the proposed methodology achieved higher accuracy compared to other feature selection methods. This methodology is a first approach to increase reproducibility, to provide robust and reliable results.

The whole-brain functional connectivity (FC) pattern obtained from resting-state functional magnetic resonance imaging data are commonly applied to study neuropsychiatric conditions such as autism spectrum disorder (ASD) by using different machine learning models. Recent studies indicate that both hyper- and hypo- aberrant ASD-associated FCs were widely distributed throughout the entire brain rather than only in some specific brain regions. Deep neural networks (DNN) with multiple hidden layers have shown the ability to systematically extract lower-to-higher level information from high dimensional data across a series of neural hidden layers, significantly improving classification accuracy for such data. In this study, a DNN with a novel feature selection method (DNN-FS) is developed for the high dimensional whole-brain resting-state FC pattern classification of ASD patients vs. typical development (TD) controls. The feature selection method is able to help the DNN generate low dimensional high-quality representations of the whole-brain FC patterns by selecting features with high discriminating power from multiple trained sparse auto-encoders. For the comparison, a DNN without the feature selection method (DNN-woFS) is developed, and both of them are tested with different architectures (i.e., with different numbers of hidden layers/nodes). Results show that the best classification accuracy of 86.36% is generated by the DNN-FS approach with 3 hidden layers and 150 hidden nodes (3/150). Remarkably, DNN-FS outperforms DNN-woFS for all architectures studied. The most significant accuracy improvement was 9.09% with the 3/150 architecture. The method also outperforms other feature selection methods, e.g., two sample t-test and elastic net. In addition to improving the classification accuracy, a Fisher's score-based biomarker identification method based on the DNN is also developed, and used to identify 32 FCs related to ASD. These FCs come from or cross different pre-defined brain networks including the default-mode, cingulo-opercular, frontal-parietal, and cerebellum. Thirteen of them are statically significant between ASD and TD groups (two sample t-test p < 0.05) while 19 of them are not. The relationship between the statically significant FCs and the corresponding ASD behavior symptoms is discussed based on the literature and clinician's expert knowledge. Meanwhile, the potential reason of obtaining 19 FCs which are not statistically significant is also provided.

10 Tips For Testing Hearing in Children With Autism

Autism spectrum disorder (ASD) is a condition that occurs in an individual, wherein it is accompanied by various symptoms such as difficulties in socializing with others. Early detection of ASD patients can assist in preventing various symptoms caused by ASD. The focus of this research is to automate the diagnosis of ASD in an individual based on the results of the autism spectrum quotient (AQ) using weighted average ensemble method. Initially, preprocessing is carried out on the dataset to ensure optimal performance of the resulting model. In the preprocessing step, the filling of missing values and feature selection occurs, where the feature selection method being utilized is p-value. The model in this research uses the weighted average ensemble method, which is the model that combines three machine learning classification algorithms. Eight classification algorithms are tested to identify the three algorithms with the best performance, namely gaussian Naïve Bayes (NB), logistic regression (LR), and random forest (RF). Following the testing, the model constructed using the weighted average ensemble method exhibits the highest performance compared to the model built using a single classification algorithm. The performance matrix used to measure the model’s performance is area under the curve (AUC)/receiver operating characteristic (ROC), with the developed model achieving an AUC/ROC value of 0.912.

Newer analytic approaches for developing predictive models provide a method of creating decision support to translate findings into practice. The aim of this study was to develop and validate a clinically interpretable predictive model for 12-month mortality risk among community-dwelling older adults. This is done by using routinely collected nursing assessment data to aide homecare nurses in identifying older adults who are at risk for decline, providing an opportunity to develop care plans that support patient and family goals for care. A retrospective secondary analysis of Medicare and Medicaid data of 635,590 Outcome and Assessment Information Set (OASIS-C) start-of-care assessments from January 1, 2012, to December 31, 2012, was linked to the Master Beneficiary Summary File (2012-2013) for date of death. The decision tree was benchmarked against gold standards for predictive modeling, logistic regression, and artificial neural network (ANN). The models underwent k-fold cross-validation and were compared using area under the curve (AUC) and other data science metrics, including Matthews correlation coefficient (MCC). Decision tree variables associated with 12-month mortality risk included OASIS items: age, (M1034) overall status, (M1800-M1890) activities of daily living total score, cancer, frailty, (M1410) oxygen, and (M2020) oral medication management. The final models had good discrimination: decision tree, AUC = .71, 95% confidence interval (CI) [.705, .712], sensitivity = .73, specificity = .58, MCC = .31; ANN, AUC = .74, 95% CI [.74, .74], sensitivity = .68, specificity = .68, MCC = .35; and logistic regression, AUC = .74, 95% CI [.735, .742], sensitivity = .64, specificity = .70, MCC = .35. The AUC and 95% CI for the decision tree are slightly less accurate than logistic regression and ANN; however, the decision tree was more accurate in detecting mortality. The OASIS data set was useful to predict 12-month mortality risk. The decision tree is an interpretable predictive model developed from routinely collected nursing data that may be incorporated into a decision support tool to identify older adults at risk for death.

Individuals with autism spectrum disorder (ASD) have difficulties in social interactions. The cognitive domains that support these interactions include perceptual decision-making, timing, and error-monitoring, which enable one to appropriately understand and react to the other individual in communicative settings. This study constitutes a comprehensive exploration of decision-making and interval timing in ASD as well as the first investigation of error-monitoring abilities of individuals with ASD regarding their performance in the corresponding domains. We found that children with ASD fared similar to typically developing (TD) children in their first-order task performance in two-alternative forced choice perceptual decision-making and temporal reproduction tasks as well as the secondary tasks (signal detection and free finger tapping tasks). Yet, they had a deficit in error-monitoring in both tasks where their accuracy did not predict their confidence ratings, which was the case for the TD group. The difference between ASD and TD groups was limited to error-monitoring performance. This study attests to a circumscribed impairment in error-monitoring in individuals with ASD, which may partially underlie their social interaction problems. This difficulty in cognitively evaluating one's own performance may also relate to theory of mind deficits reported for individuals with ASD, where they struggle in understanding the mental states and intentions of others. This novel finding holds the potential to inform effective interventions for individuals with ASD that can target this error-monitoring ability to have broad-ranging effects in multiple domains involved in communication and social interaction. Autism Res 2019, 12: 239-248 © 2018 International Society for Autism Research, Wiley Periodicals, Inc. LAY SUMMARY: Decision-making, timing, and error-monitoring are three of many abilities that underlie smooth social interactions. To date, these domains have been only investigated separately, but given their interactive role in social interactions that are impaired in ASD, we conducted the first study to investigate them together. Children with ASD were as successful as typically developing children in their task performances, but unlike them, were unaware of their errors in both decision-making and timing tasks. This deficit that is limited to error-monitoring can contribute to unraveling the unique cognitive signature of ASD and to formulating interventions with positive implications in multiple domains.

Autism spectrum disorder (ASD) is characterized by difficulties in social interaction, verbal & nonverbal communication, and restricted & repetitive behaviors. Early detection of ASD is crucial for timely intervention, yet it remains a challenging task due to the heterogeneous nature of its manifestations. Growing evidence indicates that existing screening methods, such as the Modified Checklist for Autism in Toddlers, may lack adequate precision for the early identification of children with ASD in clinical settings. Eye tracking captures subtle differences in visual attention patterns, including fixation duration, saccadic movements, and gaze direction, which are known to differ between autistic and typically developing children. The prime focus of the proposed study is to provide a comprehensive summary on the detection of autism signs by applying machine learning algorithms on eye tracking data. The eye features that hold discriminative data for the classification of autistic and typically developing children are explored in the study. The state-of-the-art methods for the detection of autism signs through eye features are presented in a coherent manner. The results of the existing methods on the classification of autistic and typically developing children are discussed in detail. The study addresses significant issues in this field, including data scarcity, class imbalance, and the interpretability of machine learning models, while also emphasizing solutions such as synthetic data production, feature selection methods, and hybrid model architectures. This review emphasizes the transformative potential of integrating eye tracking technology with machine learning to advance ASD diagnostics by synthesizing existing literature.

Autism spectrum disorder (ASD) is a multifaceted neurodevelopmental condition that affects millions of individuals globally. This study leverages the publicly available UCI Machine Learning ASD Screening Dataset, which contains behavioral questionnaire responses from children aged 4-11 years, to develop and compare multiple machine learning (ML) models for early ASD diagnosis. This study applies SHAP (SHapley Additive exPlanations) to interpret and analyze the feature importance of various attributes in the dataset. Advanced ML models like Extreme Gradient Boosting (XGBoost), Random Forest (RF), Extra Trees Classifier (ETC), Support Vector Machines (SVM), and Multilayer Perceptrons (MLPs) were implemented to evaluate and compare their performance in predicting early ASD. The developed ASD model underwent a rigorous evaluation through comprehensive performance metrics, including accuracy, precision, recall, F1-score, specificity, and Matthews correlation coefficient. Ranking-based metrics, such as area under the receiver operating characteristic curve (AUROC), provided additional insights into classification robustness, while 10-fold cross-validation ensured reliability and generalizability. Results indicate that XGBoost achieved the highest diagnostic performance (accuracy = 97.8%, AUROC = 0.99), followed closely by MLP, with RF and ETC offering balanced performance and interpretability. SHAP analysis identified key behavioral features influencing model predictions, enabling explainable AI-driven insights. These findings suggest that ML-based tools can support clinicians in identifying ASD earlier, allowing for timelier interventions in disability services, particularly in underserved populations. While results are promising, the use of secondary data and the lack of clinical validation remain limitations, highlighting the need for real-world pilot testing. These findings underscore the transformative potential of ML techniques in enhancing ASD diagnostics. Despite their promise, the results are constrained by reliance on secondary data and the absence of clinical validation, underscoring the need for real-world pilot testing. Nonetheless, incorporating these models into clinical practice could accelerate early detection, facilitate personalized interventions, and optimize resource allocation for improved ASD management outcomes.

Studies on autism spectrum disorder (ASD) have amassed substantial evidence for the role of genetics in the disease’s phenotypic manifestation. A large number of coding and non-coding variants with low penetrance likely act in a combinatorial manner to explain the variable forms of ASD. However, many of these combined interactions, both additive and epistatic, remain undefined. Coalitional game theory (CGT) is an approach that seeks to identify players (individual genetic variants or genes) who tend to improve the performance—association to a disease phenotype of interest—of any coalition (subset of co-occurring genetic variants) they join. This method has been previously applied to boost biologically informative signal from gene expression data and exome sequencing data but remains to be explored in the context of cooperativity among non-coding genomic regions. We describe our extension of previous work, highlighting non-coding chromosomal regions relevant to ASD using CGT on alteration data of 4595 fully sequenced genomes from 756 multiplex families. Genomes were encoded into binary matrices for three types of non-coding regions previously implicated in ASD and separated into ASD (case) and unaffected (control) samples. A player metric, the Shapley value, enabled determination of individual variant contributions in both sets of cohorts. A total of 30 non-coding positions were found to have significantly elevated player scores and likely represent significant contributors to the genetic coordination underlying ASD. Cross-study analyses revealed that a subset of mutated non-coding regions (all of which are in human accelerated regions (HARs)) and related genes are involved in biological pathways or behavioral outcomes known to be affected in autism, suggesting the importance of single nucleotide polymorphisms (SNPs) within HARs in ASD. These findings support the use of CGT in identifying hidden yet influential non-coding players from large-scale genomic data, to better understand the precise underpinnings of complex neurodevelopmental disorders such as autism.

The aim of the present work was to propose and implement deep neural network (DNN)-based handheld diagnosis system for more accurate diagnosis and severity assessment of individuals with autism spectrum disorder (ASD). Initially, the learning of the proposed system for ASD diagnosis was performed by implementing DNN algorithms such as a convolutional neural network (CNN) and long short-term memory (LSTM), and multilayer perceptron (MLP) with DSM-V based acquired dataset. The performance of the DNN algorithms was analyzed based on parameters viz. accuracy, loss, mean squared error (MSE), precision, recall, and area under the curve (AUC) during the training and validation process. Later, the optimum DNN algorithm, among the tested algorithms, was implemented on handheld diagnosis system (HDS) and the performance of HDS was analyzed. The stability of proposed DNN-based HDS was validated with the dataset group of 20 ASD and 20 typically developed (TD) individuals. It was observed during comparative analysis that LSTM resulted better in ASD diagnosis as compared to other artificial intelligence (AI) algorithms such as CNN and MLP since LSTM showed stabilized results achieving maximum accuracy in less consumption of epochs with minimum MSE and loss. Further, the LSTM based proposed HDS for ASD achieved optimum results with 100% accuracy in reference to DSM-V, which was validated statistically using a group of ASD and TD individuals. The use of advanced AI algorithms could play an important role in the diagnosis of ASD in today's era. Since the proposed LSTM based HDS for ASD and determination of its severity provided accurate results with maximum accuracy with reference to DSM-V criteria, the proposed HDS could be the best alternative to the manual diagnosis system for diagnosis of ASD.

Association rule has shown its usefulness in the gene expression data based disease diagnosis for its good interpretability. The large number of rules generated from the high dimensional gene expression data is one of the main challenges of its applications. In this work, we reveal that the discretization preprocessing is one of the reasons for the association rule number explosion problem. To alleviate this problem, a Regularized Gaussian Mixture Model (RGMM) is proposed to discretize the continuous gene expression data. RGMM explores both the complexity of the discretization model and the information loss of the discretization procedure, under the Minimal Description Length framework. Extensive experiments show the effectiveness of RGMM on real-life gene expression data sets.

Prediction of hepatotoxicity in humans remains an unresolved challenge. Recently, an in vitro/in silico method was established to predict blood concentrations of test compounds with an increased risk of causing human hepatotoxicity. In the present study, we addressed the question whether gene expression data can improve the quality of hepatotoxicity prediction compared to cytotoxicity analysis alone. A particular challenge is that high-dimensional gene expression data must be sum­marized into variables that allow the determination of the lowest test compound concentration that causes altered gene expression. To address this challenge, we analyzed 60 hepatotoxic and non-hepatotoxic substances in a concentration-dependent manner for cytotoxicity and expression of 3,524 probes previously reported to be influenced by hepatotoxicants. The toxicity separation index (TSI) was applied to quantify how well specific summary variables of gene expression can differen­tiate between the set of hepatotoxic and non-hepatotoxic substances. The best TSI was obtained when the lowest concentration of a test compound that led to differential expression of two genes when compared to vehicle controls was considered positive. Furthermore, the best gene expression-based summary variable was superior to cytotoxicity-based variables alone, and the combination of the best summary variables of gene expression and cytotoxicity data further improved the TSI compared to each category alone. In conclusion, the method used to derive summary variables of gene expression is critical, and the best summary variables improve the prediction of hepatotoxic substances in relation to oral doses and blood concentrations in humans.

Prediction of hepatotoxicity in humans remains an unresolved challenge. Recently, an in vitro/in silico method was established to predict blood concentrations of test compounds with an increased risk of causing human hepatotoxicity. In the present study, we addressed the question whether gene expression data can improve the quality of hepatotoxicity prediction compared to cytotoxicity analysis alone. A particular challenge is that high-dimensional gene expression data must be sum­marized into variables that allow the determination of the lowest test compound concentration that causes altered gene expression. To address this challenge, we analyzed 60 hepatotoxic and non-hepatotoxic substances in a concentration-dependent manner for cytotoxicity and expression of 3,524 probes previously reported to be influenced by hepatotoxicants. The toxicity separation index (TSI) was applied to quantify how well specific summary variables of gene expression can differen­tiate between the set of hepatotoxic and non-hepatotoxic substances. The best TSI was obtained when the lowest concentration of a test compound that led to differential expression of two genes when compared to vehicle controls was considered positive. Furthermore, the best gene expression-based summary variable was superior to cytotoxicity-based variables alone, and the combination of the best summary variables of gene expression and cytotoxicity data further improved the TSI compared to each category alone. In conclusion, the method used to derive summary variables of gene expression is critical, and the best summary variables improve the prediction of hepatotoxic substances in relation to oral doses and blood concentrations in humans. Plain language summary Liver injury is a relevant side effect of many drugs. It represents the most common cause of acute liver failure in Western countries as well as the most common reason for late-stage drug development failure or withdrawal of drugs from the market. We show that a genome-wide gene expression analysis of cultured human hepatocytes exposed to drugs can identify hepatotoxic and non-hepato­toxic substances. To achieve this, we have developed a technique to summarize all gene expression variables to determine the lowest concentration at which a test substance begins to influence gene expression. We find that the cell test must include the highest concentration of the drug that is intended to be reached in the blood of patients at therapeutic doses. This approach can reduce the number of animals needed in pre-clinical safety studies as liver-damaging drug candidates would be detected and eliminated prior to studies in rodents.

Through the eyes of autism: Eye care for these special patients

Assessment of socioeconomic and demographic risk factors for low birth weight using model-agnostic explainable ensembles.