K99-2019

1. extract image-related features

2. train model on normative dataset to predict age

3. obtain predicted ages on test set

4. compute
brain-predicted age difference (brainPAD)

5. make inference: accelerated/
delayed aging

brainPAD framework

\[brainPAD = \widehat{age}_{brain} − age_{chronological}\]

physical fitness
mortality risk in elderly participants
cognitive performance
of psychiatric disorders
mild cognitive impairment and Alzheimer’s disease

~ brainAGE: Age Gap Estimate

select voxels/ROIs

structural MRI

predicted brain age

DNA-methylation

epigenetic age

- real age

automated ML

\Delta

epigenetic age

(regularized NPDR)

~ genetic variants

epigenetic age

\Delta

Dimensions and stages of aging

Aim 3

Aim 2

Aim 2 + 3

select CpG sites

time

automated ML

brain age

\Delta

brain age

\Delta

Aim 1

Develop and evaluate NPDR

Aim 1. Identify features influencing brainPAD, optimize age prediction

Aim 2. Explore different data types of ADNI

Aim 3. Build integrative methods to create new measures of clinically abnormal aging

y_i = \beta_0 + \beta_a a_i + \epsilon_i

d_{ij}(y) = \beta_0 + \beta_a d_{ij}(a) + \epsilon_{ij}

projected distances

nearest neighbors

attributes

d_{ij}(y) = \beta_0 + \beta_a d_{ij}(a) + \epsilon_{ij}

M_a \sim d_{ij}(a|y_i \neq y_j)

H_a \sim d_{ij}(a|y_i = y_j)

discrete outcome

e.g., y \in \{0, 1\}

Supports regularization

Yes

m = 200

p = 1000

10% functionally interacting

∗ NPDR 0.05 adjusted cutoffs

https://github.com/insilico/npdr

Biological
processes

Accelerated
aging

Disease progression

\(allele_i \) carrier

\(allele_i \) non-carrier

time

composite \(\Delta\) age