Kidd, B. A. | bioRxiv Science

bioRxiv2019

Multiscale analysis of acne connects molecular subnetworks with disease status

Acne vulgaris affects millions of individuals and can lead to psychosocial impairment as well as permanent scarring. Previous studies investigating acne pathogenesis have either examined a targeted set of biological parameters in a modest-sized cohort or carried out high-throughput assays on a small number of samples. To develop a more comprehensive understanding of acne pathophysiology, we conducted an in-depth multi-omic study of 56 acne patients and 20 individuals without acne. We collected whole blood, skin punch biopsies, microbiota from skin follicles, and relevant clinical measurements to understand how multiple factors contribute to acne. We provide an integrative analysis of multi-omics data that results in a molecular network of acne. Comparisons of lesional and non-lesional skin highlighted multiple biological processes, including immune cell and inflammatory responses, response to stress, T cell activation, lipid biosynthesis, fatty acid metabolism, keratinocytes, antimicrobial activity, epithelial cell differentiation, and response to wounding, that are differentially altered in acne lesions compared to non-lesions. Our results suggest baseline differences in the skin that may predispose individuals to develop acne. These datasets and findings offer a framework for new target identification and reference for future studies.

systems biology↗

bioRxiv2017

Learning and Mapping Lyme Disease Patient Trajectories from Electronic Medical Data for Stratification of Disease Risk and Therapeutic Response

BackgroundLyme disease (LD) is an epidemic, tick-borne illness with approximately 329,000 incidences diagnosed each year in United States. Long-term use of antibiotics is associated with serious complications, including post-treatment Lyme disease syndrome (PTLDS). The landscape of comorbidities and health trajectories associated with LD and associated treatments is not fully understood. Consequently, there is an urgent need to improve clinical management of LD based on a more precise understanding of disease and patient stratification.\n\nMethodsWe used a precision medicine machine-learning approach based on high-dimensional electronic medical records (EMRs) to characterize the heterogeneous comorbidities in a LD population and develop systematic predictive models for identifying medications that influence the risk of subsequent comorbidities.\n\nFindingsWe identified 3, 16, and 17 comorbidities at broad disease categories associated with LD within 2, 5, and 10 years of diagnosis, respectively. At higher resolution of ICD-9 levels, we pinpointed specific co-morbid diseases on a timescale that matched the symptoms associated with PTLDS. We identified 7, 30, and 35 medications that influenced the risks of the reported comorbidities within 2, 5, and 10 years, respectively. These medications included six previously associated with the identified comorbidities and 29 new findings. For instance, the first-line antibiotic doxycycline exhibited a consistently protective effect for typical symptoms of LD, including backache Not Otherwise Specified (NOS) and chronic rhinitis, but consistently increased the risk of cataract NOS, tear film insufficiency NOS, and nocturia.\n\nInterpretationOur approach and findings suggest new hypotheses for precision medicine treatments regimens and drug repurposing opportunities tailored to the phenotypic profiles of LD patients.\n\nFundingThe Steven & Alexandra Cohen Foundation

bioinformatics↗