Toward personalized care for insomnia in the US Army: a machine learning model to predict response to cognitive behavioral therapy for insomnia.

STUDY OBJECTIVES: The standard of care for military personnel with insomnia is cognitive behavioral therapy for insomnia (CBT-I). However, only a minority seeking insomnia treatment receive CBT-I, and little reliable guidance exists to identify those most likely to respond. As a step toward personalized care, we present results of a machine learning (ML) model to predict CBT-I response. METHODS: Administrative data were examined for n=1,449 nondeployed US Army soldiers treated for insomnia with CBT-I who had moderate-severe baseline Insomnia Severity Index (ISI) scores and completed one or more follow-up ISIs 6-12 weeks after baseline. An ensemble ML model was developed in a 70% training sample to predict clinically significant ISI improvement (reduction of at least two standard deviations on the baseline ISI distribution). Predictors included a wide range of military administrative and baseline clinical variables. Model accuracy was evaluated in the remaining 30% test sample. RESULTS: 19.8% of patients had clinically significant ISI improvement. Model AU-ROC (SE) was 0.60 (0.03). The 20% of test sample patients with highest probabilities of improvement were twice as likely to have clinically significant improvement as the remaining 80% (36.5% versus 15.7%; χ21=9.2, p=.002). Nearly 85% of prediction accuracy was due to ten variables, the most important of which were baseline insomnia severity and baseline suicidal ideation. CONCLUSIONS: Pending replication, the model could be used as part of a patient-centered decision-making process for insomnia treatment. Parallel models will be needed for alternative treatments before such a system is of optimal value.

Toward personalized care for insomnia in the US Army: development of a machine-learning model to predict response to pharmacotherapy.

STUDY OBJECTIVES: Although many military personnel with insomnia are treated with prescription medication, little reliable guidance exists to identify patients most likely to respond. As a first step toward personalized care for insomnia, we present results of a machine-learning model to predict response to insomnia medication. METHODS: The sample comprised n = 4,738 nondeployed US Army soldiers treated with insomnia medication and followed 6-12 weeks after initiating treatment. All patients had moderate-severe baseline scores on the Insomnia Severity Index (ISI) and completed 1 or more follow-up ISIs 6-12 weeks after baseline. An ensemble machine-learning model was developed in a 70% training sample to predict clinically significant ISI improvement, defined as reduction of at least 2 standard deviations on the baseline ISI distribution. Predictors included a wide range of military administrative and baseline clinical variables. Model accuracy was evaluated in the remaining 30% test sample. RESULTS: 21.3% of patients had clinically significant ISI improvement. Model test sample area under the receiver operating characteristic curve (standard error) was 0.63 (0.02). Among the 30% of patients with the highest predicted probabilities of improvement, 32.5.% had clinically significant symptom improvement vs 16.6% in the 70% sample predicted to be least likely to improve (χ21 = 37.1, P < .001). More than 75% of prediction accuracy was due to 10 variables, the most important of which was baseline insomnia severity. CONCLUSIONS: Pending replication, the model could be used as part of a patient-centered decision-making process for insomnia treatment, but parallel models will be needed for alternative treatments before such a system is of optimal value. CITATION: Gabbay FH, Wynn GH, Georg MW, et al. Toward personalized care for insomnia in the US Army: development of a machine-learning model to predict response to pharmacotherapy. J Clin Sleep Med. 2023;19(8):1399-1410.

The relationship between DNA methylation, genetic and expression inter-individual variation in untransformed human fibroblasts.

BACKGROUND: DNA methylation plays an essential role in the regulation of gene expression. While its presence near the transcription start site of a gene has been associated with reduced expression, the variation in methylation levels across individuals, its environmental or genetic causes, and its association with gene expression remain poorly understood. RESULTS: We report the joint analysis of sequence variants, gene expression and DNA methylation in primary fibroblast samples derived from a set of 62 unrelated individuals. Approximately 2% of the most variable CpG sites are mappable in cis to sequence variation, usually within 5 kb. Via eQTL analysis with microarray data combined with mapping of allelic expression regions, we obtained a set of 2,770 regions mappable in cis to sequence variation. In 9.5% of these expressed regions, an associated SNP was also a methylation QTL. Methylation and gene expression are often correlated without direct discernible involvement of sequence variation, but not always in the expected direction of negative for promoter CpGs and positive for gene-body CpGs. Population-level correlation between methylation and expression is strongest in a subset of developmentally significant genes, including all four HOX clusters. The presence and sign of this correlation are best predicted using specific chromatin marks rather than position of the CpG site with respect to the gene. CONCLUSIONS: Our results indicate a wide variety of relationships between gene expression, DNA methylation and sequence variation in untransformed adult human fibroblasts, with considerable involvement of chromatin features and some discernible involvement of sequence variation.

Computational analysis of whole-genome differential allelic expression data in human.

Allelic imbalance (AI) is a phenomenon where the two alleles of a given gene are expressed at different levels in a given cell, either because of epigenetic inactivation of one of the two alleles, or because of genetic variation in regulatory regions. Recently, Bing et al. have described the use of genotyping arrays to assay AI at a high resolution (approximately 750,000 SNPs across the autosomes). In this paper, we investigate computational approaches to analyze this data and identify genomic regions with AI in an unbiased and robust statistical manner. We propose two families of approaches: (i) a statistical approach based on z-score computations, and (ii) a family of machine learning approaches based on Hidden Markov Models. Each method is evaluated using previously published experimental data sets as well as with permutation testing. When applied to whole genome data from 53 HapMap samples, our approaches reveal that allelic imbalance is widespread (most expressed genes show evidence of AI in at least one of our 53 samples) and that most AI regions in a given individual are also found in at least a few other individuals. While many AI regions identified in the genome correspond to known protein-coding transcripts, others overlap with recently discovered long non-coding RNAs. We also observe that genomic regions with AI not only include complete transcripts with consistent differential expression levels, but also more complex patterns of allelic expression such as alternative promoters and alternative 3' end. The approaches developed not only shed light on the incidence and mechanisms of allelic expression, but will also help towards mapping the genetic causes of allelic expression and identify cases where this variation may be linked to diseases.

PSORTb 3.0: improved protein subcellular localization prediction with refined localization subcategories and predictive capabilities for all prokaryotes.

MOTIVATION: PSORTb has remained the most precise bacterial protein subcellular localization (SCL) predictor since it was first made available in 2003. However, the recall needs to be improved and no accurate SCL predictors yet make predictions for archaea, nor differentiate important localization subcategories, such as proteins targeted to a host cell or bacterial hyperstructures/organelles. Such improvements should preferably be encompassed in a freely available web-based predictor that can also be used as a standalone program. RESULTS: We developed PSORTb version 3.0 with improved recall, higher proteome-scale prediction coverage, and new refined localization subcategories. It is the first SCL predictor specifically geared for all prokaryotes, including archaea and bacteria with atypical membrane/cell wall topologies. It features an improved standalone program, with a new batch results delivery system complementing its web interface. We evaluated the most accurate SCL predictors using 5-fold cross validation plus we performed an independent proteomics analysis, showing that PSORTb 3.0 is the most accurate but can benefit from being complemented by Proteome Analyst predictions. AVAILABILITY: http://www.psort.org/psortb (download open source software or use the web interface). CONTACT: psort-mail@sfu.ca SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Benchmarking HIPAA compliance.

One of the nation's largest academic medical centers is benchmarking its operations using internally developed software to improve privacy/confidentiality of protected health information (PHI) and to enhance data security to comply with HIPAA regulations. It is also coordinating the development of a web-based interactive product that can help hospitals, physician practices, and managed care organizations measure their compliance with HIPAA regulations.