fairfolio
Saved Projects

ALZCan: Machine Learning Based Multimodal Neuroimaging Genetics Framework to Predict Future Onset of Alzheimer's Disease

CSEF · 2017 Computational Systems & Analysis Fourth Award

Pravin Ravishanker · Bellarmine College Preparatory, San Jose, Santa Clara County

Overview

Objectives/Goals To create an accurate diagnostic and prognostic machine learning based methodology for Alzheimer's early detection by using polygenic risk scores constructed from single nucleotide polymorphism (SNP) genotype data; cerebrospinal fluid (CSF) levels; cognitive tests; demographic info; 3 neuroimaging modalities (678 Resting State Functional MRI (rs-fMRI), 427 Fluorodeoxyglucose PET (FDG-PET), 407 Florbetapir PET brain scans). Methods/Materials Using R programming language and data from Alzheimer's Disease Neuroimaging Initiative (ADNI), I considered 23 neuroimaging and CSF phenotypes linked to AD pathology: 3 CSF protein levels; FDG-PET cerebral metabolic activity in 5 brain regions; Florbetapir-PET beta-amyloid plaque levels in 7 regions; and 8 rs-fMRI functional network connectivity (FNC) measures (3 ROI-to-Voxel Ratio + 5 Graph-theory based). Graph-theoretic FNC metrics were extracted from each rs-fMRI scan using ICA, signal Cross-Correlation, and graph-rendering algorithms. Using GLMs, Genome-Wide Association Analyses evaluated 14 million+ interactions between 608,586 SNP genetic variants (393 participant genotypes) and the aforesaid 23 AD phenotypes. 23 polygenic risk scores per subject were computed by combining weighted additive effect of each subject's multiple SNP variants, with association results determining each SNP's weight in risk scores. Results Based on a subject's 23 polygenic risk scores, demographic info, and cognitive scores; gradient boosting machine learning could differentiate between Healthy Control, Mild Cognitive Impairment (MCI), Alzheimer's (AD) with diagnostic accuracy of 98.10%; and predict onset of AD and MCI 6, 12, 24, 36 months into future with prognostic accuracies of 91.89%, 91.72%, 85.38%, 70.67% respectively. Polygenic risk scores constructed from top genetic variants affecting 1) Posterior Cingulate/Left Angular Gyrus metabolic activity; 2) Precuneus beta-amyloid plaque levels; 3) rs-fMRI FNC as measured through Default Mode Network ROI-to-Voxel ratio and Transitivity; 4) CSF phosphorylated tau levels were most capable of predicting MCI and AD status in the future. Conclusions/Discussion With power of polygenic risk scores for risk prediction/intervention/personalized medicine and machine learning's promise to discover relationships amongst massive genomic/neuroimaging datasets, we can precisely pinpoint Alzheimer's future onset and prevent irreversible brain damage.

Summary statement

I created a novel machine learning framework for Alzheimer's and Mild Cognitive Impairment's early diagnosis and prognosis, fusing signal processing, graph theory, neuroimaging, genome-wide association analyses, and polygenic risk scoring.

Help received

I thank the Alzheimer's Disease Neuroimaging Initiative (international effort tracking AD biomarkers) and its investigators for the vital data repository. I am grateful to all who have supported me throughout my 4-year progressive computational neuroscience research on Alzheimer's prediction.

Awards (1)

Competition history

  • CSEF 2017 Computational Systems & Analysis · Entry S0827

Resources

Related projects

Closest projects by meaning, across every fair and year in the corpus.

Browse more like this

Source: California Science & Engineering Fair public projects

Save projects to your library

Sign in with Google to keep track of projects you find interesting, organized into folders. An account also raises your daily allowance for “Has this been done?”, and lets you create a key for the MCP server with a much higher limit than anonymous use. Browsing stays public.

Continue with Google