University of Michigan

Brian D Athey

Alex S Ade

Gerald A Higgins

Alexandr Kalinin

James S Burns

Narathip Reamaroon

Ari Allyn-Feuer

Brian Pauls

David S Dilworth

Ivaylo Dinov

Methods comprising an integrated, multiscale artificial intelligence-based system that reconstructs drug-specific pharmacogenomic networks and their constituent functional sub-networks are described. The system uses features of the functional topology of the three-dimensional architecture of drug-modulated spatial contacts in chromatin space. Discovery of a drug pharmacogenomic network is made through the selection of candidate SNPs by imputation, determination of the predicted causality of the SNPs using machine learning and deep learning, use of the causal SNPs to probe the spatial genome as determined by chromosome conformation capture analysis, combining targeted genes controlled by the same cell and tissue-specific enhancers, and reconstruction of the pharmacogenomic network using diverse data sources and metrics based on the results of genome-wide association studies. Knowledge-based segmentation methods are used to deconstruct the pharmacogenomic network into its constituent efficacy and adverse event sub-networks for applications in clinical decision support, drug re-purposing, and in silico drug discovery.

Methods and system for the reconstruction of drug response and disease networks and uses thereof

Methods for identifying patients diagnosed with treatment resistant or refractory depression, pain or other clinical indications who are eligible to receive N-methyl-D-aspartate receptor antagonist, glycine receptor beta (GLRB) modulator, or α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid receptor (AMPAR)-based therapies to include determining the appropriate medication, an optimal dose for each patient, and determining which patients are not eligible to receive the therapy. The pharmacogenomic clinical decision support assays include targeted single nucleotide polymorphisms and clinical values or a combination of targeted single nucleotide polymorphisms, targeted ketamine-specific expansion and contraction of topologically associated domains, and clinical values. The methods described herein allow for a more effective determination of which patients will experience drug efficacy and which patients will experience adverse drug events. The methods provide personalized patient recommendations for dose, the frequency of medication administration, and recommendations on drug choice.

Pharmacogenomic decision support for modulators of the NMDA, glycine, and AMPA receptors

To analyze spatial organization of chromatin a computing device may compile genomic element contacts or reads into variable size bins using a binary search tree. The bins may be selected to each represent a different cutsite increment or functional element within a genome, such as a gene, TAD, chromatin state segment, loop domain, chromatin domain, etc. Two sets of bins are selected to generate a squared genome matrix of bin pairs, where each set represent an axis of the matrix. Then a normalization method is applied to the interaction frequencies for the bin pairs having variable size and/or shape to generate normalized interaction frequencies for each bin pair. The normalized interaction frequencies may be used to identify bin pairs having enriched and depleted contacts for a variety of analyses, including the detection of target genes of genomic variants, as well as genome wide analysis of contacts.

Method and apparatus for analysis of chromatin interaction data

The ability to automate the processes of specimen collection, image acquisition, data pre-processing, computation of derived biomarkers, modeling, classification and analysis can significantly impact clinical decision-making and fundamental investigation of cell deformation. This disclosure combine 3D cell nuclear shape modeling by robust smooth surface reconstruction and extraction of shape morphometry measure into a highly parallel pipeline workflow protocol for end-to-end morphological analysis of thousands of nuclei and nucleoli in 3D. This approach allows efficient and informative evaluation of cell shapes in the imaging data and represents a reproducible technique that can be validated, modified, and repurposed by the biomedical community. This facilitates result reproducibility, collaborative method validation, and broad knowledge dissemination.

Three-dimensional cell and tissue image analysis for cellular and sub-cellular morphological modeling and classification

For patients who exhibit or may exhibit primary or comorbid disease, pharmacological phenotypes may be predicted through the collection of panomic data over a period of time. A machine learning engine may generate a statistical model based on training data from training patients to predict pharmacological phenotypes, including drug response and dosing, drug adverse events, disease and comorbid disease risk, drug-gene, drug-drug, and polypharmacy interactions. Then the model may be applied to data for new patients to predict their pharmacological phenotypes, and enable decision making in clinical and research contexts, including drug selection and dosage, changes in drug regimens, polypharmacy optimization, monitoring, etc., to benefit from additional predictive power, resulting in adverse event and substance abuse avoidance, improved drug response, better patient outcomes, lower treatment costs, public health benefits, and increases in the effectiveness of research in pharmacology and other biomedical fields.

Individual and cohort pharmacological phenotype prediction platform

For patients who exhibit or may exhibit primary or comorbid disease, pharmacological phenotypes may be predicted through the collection of panomic, physiomic, environmental, sociomic, demographic, and outcome phenotype data over a period of time. A machine learning engine may generate a statistical model based on training data from training patients to predict pharmacological phenotypes, including drug response and dosing, drug adverse events, disease and comorbid disease risk, drug-gene, drug-drug, and polypharmacy interactions. Then the model may be applied to data for new patients to predict their pharmacological phenotypes, and enable decision making in clinical and research contexts, including drug selection and dosage, changes in drug regimens, polypharmacy optimization, monitoring, etc., to benefit from additional predictive power, resulting in adverse event and substance abuse avoidance, improved drug response, better patient outcomes, lower treatment costs, public health benefits, and increases in the effectiveness of research in pharmacology and other biomedical fields.

Growing community of inventors

Ann Arbor, MI, United States of America

Brian D Athey