\( \def\RR{\bf R} \def\real{\mathbb{R}} \def\bold#1{\bf #1} \def\d{\mbox{Cord}} \def\hd{\widehat \mbox{Cord}} \DeclareMathOperator{\cov}{cov} \DeclareMathOperator{\var}{var} \DeclareMathOperator{\cor}{cor} \newcommand{\ac}[1]{\left\{#1\right\}} \DeclareMathOperator{\Ex}{\mathbb{E}} \DeclareMathOperator{\diag}{diag} \)
**Network Modeling of HIV and Brain
**

### Xi (Rossi) **LUO**

## Why Important for HIV and Neuroscience

## Ex 1: Genes and HIV Brains

## Method Comparision

## Results

Recovered Gene Networks
Our (red line) prediction vs other AIs (higher accuracy better)
## Ex 2: HIV + Molecules + Imaging

## Ex 3: Predicting Diseases

# Demo

## Summary

# Thank you!

## Comments? Questions?

## Slides at:
Big Complex Data**.com**

**Brown University**

Department of Biostatistics

Center for Statistical Sciences

Computation in Brain and Mind

Brown Institute for Brain Science

Brown Data Science Initiative

The ABCD Research Group

**2017 CFAR Forum**

May 19, 2017

Funding: NSF/DMS (BD2K) 1557467; NIH R01EB022911, P20GM103645, P01AA019072, P30AI042853

**Overview**: machine learning, Bayesian, probability/matrix theory, optimization, large-scale computing**Idea**: integrate techniques from CS, Biology, Math, and Stat to solve some complex problems

- Big data
- genes, cytokines/chemokines, neuroimaging

- Complex data
- Multiple data domains, visits
- Comorbidities, behavior

- Big and complex
- Link big data with complex outcomes

- Gene expression arrays on post-mortem brain tissues Borjabad et al, 2011
- Tens of thousands genes
- Outcome: controls, treated HAND, untreated HAND

- Data: publicly available from NIH GEO GSE28160
- R package also publicly available from CRAN
- Follow NIH's call for reproducible research

- Question: are there relationships between expressions and HIV treatment?

- Massive testing
- T tests for each gene vs HAND treatment (Yes/No)
- Drawbacks: far from biology, no validation, sample size, multiple testing

- Network modeling
- Goal: how all genes together as networks related to treatment
- Need computational and mathematical tools to find gene networks

- Black-box machine learning and artificial intelligence
- Feed genes into predictive models
- Examples: random forests, support vector machine
- Test prediction performance on an independent sample

- With ARCH investigators, R Cohen and others
- Data:
- Baseline covariates
- Many cytokine and chemokine markers
- Many imaging outcomes in multiple brain regions at multiple visits
- Magnetic resonance spectroscopy
- Regions: front, middle, ...

- How (some) markers influence imaging (trends)?

Our method removed many insignificant combinations automatically

- Other comorbidities: T2D or CVD
- Recent collaboration with Drs Ingalls and Ray (BU) on HIV

- Make these computational tools widely available
- Use for behavioral interventions?

- Network modeling
- Complex models for complex diseases

- New translational tools and technologies
- Much more to be done
- Thanks to CFAR and ARCH!
- Especially, my mentor Joe Hogan and many other colleagues
- Team: R Cohen, B Navia, and students