Learning the Regulatory Code of Alzheimer's Disease Genomes

Our overarching objective is to apply machine learning techniques to predict and interpret the functional effects of genetic variants including Single Nucleotide Variants (SNVs), indels and Structural Variants (SVs) from AD WGS data at the levels of DNA regulation and RNA processing, and link these effects directly to pathways and network context. We will leverage WGS generated by the ADSP and others together with harmonized endophenotypes and clinical data, multi-omics data from the AMP-AD, functional genomics data from Roadmap Epigenomics, PsychENCODE and GTEx Projects, and microglia and monocytes specific transcriptomic and single-cell RNA-seq data sets. Our central hypothesis is that many AD-associated genetic risk or protective variants influence pre- and post-transcriptional gene regulation, resulting in changes to gene expression and cellular pathways/networks, and ultimately contribute to protein aggregation in AD. The objective of this aim is to leverage deep-learning-based models capable of predicting functional effects of genomic variants on pre- and post-transcriptional gene regulation. We will train existing and novel sequence-based deep learning models of epigenomic state and RNA regulation and processing specific to AD-relevant cell types and states. in silico mutagenesis under these trained models will be used to calculate functional impact “delta scores” for every SNV, indel and structural variants (SV) detected from AD WGS. We will use these delta scores to empower non-coding rare variant tests of association with AD at the regulatory region, gene and pathway levels. We will conduct functional fine-mapping through the integration of (i) the CNN delta scores (ii-iii) expression and splicing quantitative trait loci (eQTL and sQTL), (iv) AD endophenotypes and (v) multi-ethnic AD WGS data. We will use probabilistic ML methods, combined with cell-type-specific and single-cell RNA-seq datasets, to build gene regulatory networks. This NIH funded project is a close collaboration with Dr. Towfique Raj at Mount Sinai Medical School.

Despite decades of research and enormous investment, no disease-modifying treatment is available for Alzheimer’s disease (AD). Combining population-scale data collection, human genetics and machine learning provide a way forward to uncover and characterize new causal cellular processes involved in AD. Effectively integrating diverse genomic data to better understand AD represents a substantial computational challenge, both in terms of data scale and analysis complexity. We will train machine learning models to predict epigenomic signals from the genomic sequences to estimate the functional impact of any genetic variant. These analyses will highlight variants and genes involved in AD. However, genes do not operate in a vacuum so robust machine learning will be used to learn cell-type and disease- specific networks. Such pathways will be prime candidates for future functional and therapeutic studies of AD.