A Prototype of Brain Network Simulator for Spatiotemporal Dynamics of Alzheimer’s Disease

A Prototype of Brain Network Simulator for Spatiotemporal Dynamics of Alzheimer’s Disease

一個模擬阿茲海默症之時空動態的腦網路模擬器原型

Speaker : Jimmy Lu 盧松筠Advisor : Hsing Mei 梅　興

Web Computing Laboratory (WECO Lab)

Computer Science and Information Engineering Department

Fu Jen Catholic University

04/18/2023 WECO Lab http://www.weco.net 2

Outline

• Introduction• Motivation• Background and Related Work• The Brain Network Simulator

– Design Concepts and Development Approaches

• Alzheimer’s Disease– Three Different Models– The Proposed Spatiotemporal Model of Alzheimer’s Disease

• Implementation and Demo• Conclusion and Future Work


Introduction

• It’s the Decade of Brain!• NIH Blueprint for Neuroscience Research

– Grand Challenges• the connectivity of the adult human brain• targeted therapy development for neurological diseases

• Collaborative Works In the Multi-disciplinary Research Field– Computer Science plays a key role

• Brain Network Simulator– Modeling structural and functional dynamics of the human brain– Apply to different cases (brain functions, diseases, cognition, behavior)– Keep evolving– education, research, diagnosis, personal health care, etc.


Motivation

• Few studies by similar approach– Because the issue is extremely complex– But we’d loved to be the pioneer

• The start of the Human Connectome Project– Connection map will be the foundation of brain network simulator

• The human brain is a large network– In IT research field, we have experience on real network analysis– The experiences can be inspirations for study brain networks

• We believe simulation is the trend in the future of brain science studies


Background and Related Work



• Brain informatics– An emerging interdisciplinary

research field– Human Information Processing

System (HIPS)– Technology in web

intelligence, especially in deep web intelligence, such as data mining, machine learning, and social network analysis, helps studies of brain science Brain Informatics

Cognitive Science

Neuroscience

Web Intelligence

Deep Web Intelligence



• The Human Connectome Project– Comprehensive map of neural connections in the human brain will be

the foundation of studies of brain science• The-state-of-art neuroimaging technology• Macroscopic connectomes

• Brain Networks– by Connection Type

• Anatomical connectivity• Functional connectivity• Effective connectivity

– by Functionality• Thalamocortical Motifs• Polysynaptic Loop Structure• Diffuse Ascending Projections


Basic Brain Networks

(a) Thalamocortical Motif

(b) Polysynaptic Loop Structure (c) Diffuse Ascending Projections

GPe – External Global PallidusGPi – Internal Global PallidusSTN – Subthalamic NucleusSNc – Substantia Nigra CompactaSNr – Substantia Nigra Retuculata

DA – Dopamine5-HT – SerotoninAch – Acetylcholine



• Complex Network Analysis– Graph theory– targets: real life network– including brain networks– structure-function mapping

• Alzheimer’s Disease– the most common dementia– unknown causes, incurable, degenerative, and terminal disease– four stages shows different patterns of impairments and symptoms on

cognitive functions– lasts a long period of time

shortest path cluster

modules



• Current Status of Brain Simulator

IBM’s C2 Blue BrainProposed Brain

Simulator

PerspectiveNeuron-LevelMicroscopic

Neuron-LevelMicroscopic

Brain-LevelMacroscopic

Basic Component Neuron Neuron Nuclei, Region, Tracts

Connection Synapse SynapseCommunication

Pathway

Communication Electrical Signal Electrical Signal Protocol Data Unit

Architecture P2P Network layered Architecture layered Architecture

Focus Area Cortex Neocortical column Whole Brain

ComputationSupercomputer

Blue GeneSupercomputer

Blue GeneCloud Computing

Environment

Granularity Fine-grained Fine-grained Coarse-grained

Approach HardwareHardware and

SoftwareSoftware


The Brain Network Simulator



• Design Concepts and Approaches– Architecture

• Comparison between brain networks and the Internet• Layered architecture inspired by the Internet

– Data Structure• Graph Structure: node and edge• Brain Components

– thalamus– hippocampus– acetylcholine

– Workflow– Development Approach

• Case-based incremental delivery











• Internet vs. Brain NetworksInternet Human Brain

Scale Billions of unit elements 1011 unit elements

Layered structure OSI modelAnatomical structure, network

overlays, and functional outputs

Mechanisms of fault toleranceError correction, recomputation

of routing pathsDegeneracy mechanism,

replaceable functional areas

Properties of complex networksMotif, communities, hubs,

shortest pathway, etc.Motif, communities, hubs,

shortest pathway, etc.

Capability of an unit element Versatile Specific

Network topologyDynamic (by leaving or joining

of computers)Dynamic (by learning, aging, or developmental processes)


Layered Architecture of Brain Simulator

Brain Connectivity Layer

Processing Layer

Application Layer(Behavior/Disease/Cognitive Functions)

Causal Layer(Overlays)

Time Scale

Polysynaptic Loops Diffuse Ascending ProjectionThalamocortical Motif

ShortTerm

LongTerm

Cognitive System

Decision Making

Resting State

Sleep Aging

Neural Darwin Selection

Network Development

Model

Network Damage Model

Brain Disease

Brain Disease Models

………

……

……

ReasoningSleep Switch

Model










Connections are maintained by a sparse matrix to optimize memory usage













A Workflow Scenario of Brain Network Simulator

Input Data

Instantiate Brain Components to Create Brain Anatomical Network

Signal Filtering, Image Normalization, Transformation, etc.

Data Preprocessing

Extract Required Information

3D Brain Network Rendering

time

Apply Theoretical Model for Simulation Network Analysis

Research or experiment results










Case-based Incremental Delivery

Case Study and Analysis

Layered Architecture Extending and Refactoring

Brain Components Extending and Refactoring

Build Theoretical Models

Evaluate Theoretical Models

Existing C

ases

Evolved Brain Simulator

Feedback

Cases Integration

Research or Experiment

Results

Personalized Medical data

New Cases

Model Pool


Spatiotemporal dynamics of Alzheimer’s Disease



• Three different models– Neuropathological stageing of Alzheimer-related changes

• Describe global pattern of lesions caused by Alzheimer’s disease• Lesions: distribution of amyloid and neurofibrillary changes

– Network Damage Model• Intentional attack on the node with highest degree• Observed in the brain affected by Alzheimer’s disease• Focus on fragments after attack

– Treatment• Based on cholingeric hypothesis• Needs to find out the cholingeric pathways

• A spatiotemporal model of Alzheimer’s Disease– A combination of three with temporal parameter added in

SIMULATE-ALZHEIMER’S-DISEASE(time t, network $s)1 while time(t) < tend

2 affectedRegions[] GLOBAL-PATTERN-OF-LESIONS(t)4 for each region r affectedRegions[]5 do targetNodes[] CHOOSE-TARGET-NODES(t, r)6 for each node n targetNodes[]7 do compute the decreased number of neurons within n8 do update s9 for each edge e that connects to n10 do compute the decreased number of connections11 do re-compute the weight w of edge e12 do update s










Basal Portion of Frontal Lobe

Basal Portion of Limbic Lobe

Basal Portion of Occipital Lobe

Isocortex Association Area

Isocortical Areas(including the belt fields and primary areas)

Stage I

Stage II

Stage III

Distribution Pattern of Amyloid Deposits


Transentorhinal Region

Limbic Area(involve the entorhinal and transentorhinal layer Pre-α)

Isorcortex

Stage III & IV

Stage I & II

Stage V & VI

Distribution Pattern of Neurofibrillary Tangles and Neuropil Threads









Hub

Remove Hubs

A Cluster Three Cluster

Network Damage Model

𝑞𝑘=𝜃 (𝑘𝑚𝑎𝑥−𝑘 )={1𝑖𝑓 𝑘≤𝑘𝑚𝑎𝑥

0 𝑖𝑓 𝑘>𝑘𝑚𝑎𝑥

Where is the probability a node will be occupied, is the is the Heaviside step function, is the degree threshold, is the degree of a node

• It has been applied to some studies of Alzheimer’s disease









Neurochemical Changes in Alzheimer’s Disease

𝑡𝑎𝑢⇌𝑡𝑎𝑢p APP

Postsynaptic NeuronPresynaptic Neuron Synapatic Cleft

Ca2+ACh

ChAT

Acetyl-CoA

Choline

AChE Inhibitor

Nerve Impulse

Vesicles

AChE

ChAT – Choline AcetyltransferaseACh – AcetylcholineAChE – AcetylcholinesteraseAPP – Amyloid Precursor Protein

ACh Receptor


Cholinergic Pathways

Ch1Ch2

Ch3Ch4

Ch5 Ch6

neocortex

hippocampus

cingulate

retrosplenia

olfactory bulb

visual areathalamus

deep cerebellar nuclei

amygdala

Ch1 – medial septumCh2 – vertical limb nucleusCh3 – horizontal limb nucleusCh4 – nucleus basalisCh5 – pedunculopontine nucleusCh6 – lateral dorsal tegmental nucleus









5K

4K

1K

3.6

2.0

1.2

Local View

Global View

Global and Local Views of Alzheimer’s Brain


Where is the Heaviside step function, represents a threshold of degree, is the maximum degree in a local region, is the degree of node , is the start point of the simulation, is a period of time that controls the duration of an attack

,

𝑙𝑒𝑡 𝑘𝑡𝑎𝑟𝑔𝑒𝑡= 𝑓 (𝑡 )=𝑘𝑚𝑎𝑥 − ⌈𝑡−𝑡 0

𝑝⌉

node in the network at time ,


target node in the network, the total decreased number of neurons at time is

where is the decreased number of neurons, is the speed of neuron deaths, is the speed of neuron deaths at time , is the constant speed of neuron deaths, is the amount of acetylcholine at time ,


edge in the network with source node and target node , the weight of at time is

𝑾 (𝒕𝒏 )= 𝜷𝜶

×𝑪 (𝒕𝒏)𝟏𝟎𝟒 ×

𝟏𝒍

where are coefficients to determine the ratio between and , notice that , is the number of connections that compose at time , is the length of

𝑪 (𝒕𝒏 )={ 𝑵𝑺 (𝒕𝒏 ) ×𝟏𝟎𝟒×𝒚

𝒙+𝒚×

𝑵𝑻 (𝒕𝒏)

∑𝒊=𝟎

𝒚

𝑵 𝑻 𝒊(𝒕𝒏)

𝒊𝒇 𝒏=𝟎

𝑪 (𝒕𝒏−𝟏) × {𝟏−∆𝒏𝑺

𝑵𝑺 (𝒕𝒏−𝟏 )𝒊𝒇 ∆𝒏𝑺≥ ∆𝒏𝑻

𝟏−∆𝒏𝑻

𝑵𝑻 (𝒕𝒏−𝟏 )𝒊𝒇 ∆𝒏𝑺<∆𝒏𝑻

𝒊𝒇 𝒏>𝟎

where and are the number of inlinks and outlinks respectively, and are the decreased number of and respectively from to


Steps of Brain Network Simulation of Alzheimer’s Disease

5

3 1

1.875

1.66

0.33

0.6250.33

t = 0

3

3 1

1.125

1

0.33

0.3750.33

t = 1

2

1 1

0.375

0.33

0.11

0.250.13

t = 2

ACh

Assume that are all equal to 1, is 2 per unit time, and is a factor of 2, then the dynamics of weights are as follow:


Demo


Conclusion

• Brain simulation is the trend in the future of brain science studies

• Try to design a brain network simulator– Layered architecture inspired by network comparison– Brain components– Workflow– Development approach


• A spatiotemporal model of Alzheimer’s disease• A prototype of brain network simulator


Future Work

• Brain network simulator development– Brain components refinement– Input data and data preprocessing– Network analysis– Distributed computing

• Evolved brain network simulator– Add more cases into the brain network simulator– Ex: research result or experiment data of sleep

• Usage– Research– Diagnosis– Personal healthcare


Q&A


Thanks For Listening!