amia tbi 2010_pmi_luciano.ppt

Translational Medicine:!Using Systems of Differential Equations to Identify Patterns in Symptom Remission in Response to Treatment and the !

Underlying Dynamics of their Interactions!

Joanne  S.  Luciano,  Ph.D.  Predictive  Medicine,  Inc.  

Belmont,  MA  

2010  AMIA  Summit  on  Translational  Bioinformatics  Parc  55  Hotel  San  Francisco  

San  Francisco,  California,    USA  March  11,  2010  

Predictive Medicine, Inc. © 2010! 2!

Take Home Messages !!!A neural network model is capable of predicting and describing recovery patterns in depression!!Recovery patterns differ treatment!

•  Cognitive Behavioural Therapy!» is sequential!

•  Desipramine!» is simultaneous and delayed

!!

Predictive Medicine, Inc. © 2010! 3!

Overview!•  Why we did this work - to improve quality of life for millions

of people suffering from depression!•  How we did it - used differential equations (“neural

network”) to model and compare response to different antidepressant treatments!

•  What we found - different response patterns for the two treatments - the order and timing of improvement of symptoms were different!

•  What we think it means - improvement in selection of treatment thereby reducing unnecessary costs and suffering. Potentially saving lives!

Predictive Medicine, Inc. © 2010! 4!

Overview!•  Why we did this work - to improve quality of

life for millions of people suffering from depression!

•  How we did it - used differential equations (“neural network”) to model and compare response to different antidepressant treatments!

•  What we found - different response patterns for the two treatments - the order and timing of improvement of symptoms were different!

•  What we think it means - improvement in selection of treatment thereby reducing unnecessary costs and suffering. Potentially saving lives!

Predictive Medicine, Inc. © 2010! 5!

Translational Medicine!•  Rapid transformation of laboratory findings into

clinically focused applications !•  ʻFrom bench to bedside and backʼ!

Predictive Medicine, Inc. © 2010! 6!

Depression is a BIG problem!Characterized by persistent and pathological sadness,

dejection, and melancholy!Prevalence (US)!!6% year (18 million)!!16% experience it in their lifetime!

Cost !!44 Billion (1990)!

Impact!!1% Improvement means (180, 000 people helped)!!1% Improvement means (440 million in savings)!

Predictive Medicine, Inc. © 2010! 7!

The  Economic  Burden  of  Depression  

Source: The Healthy Thinking Initiative!

http://www.preventingdepression.com/costs.htm

Depression is the highest of the health care cost for business

Predictive Medicine, Inc. © 2010!

Depression is a BIG Problem!

Predictive Medicine, Inc. © 2010!

Treatment Choice Vague!

Predictive Medicine, Inc. © 2010! 10!

Overview!•  Why we did this work - to improve quality of life for millions

of people suffering from depression!•  How we did it - used differential equations

(“neural network”) to model and compare response to different antidepressant treatments!

•  What we found - different response patterns for the two treatments - the order and timing of improvement of symptoms were different!

•  What we think it means - improvement in selection of treatment thereby reducing unnecessary costs and suffering. Potentially saving lives!

Predictive Medicine, Inc. © 2010! 11!

   Research Goals!

Illuminate recovery course

Predictive Medicine, Inc. © 2010! 12!

Today’s  talk:  Response  to  treatment  

Treatment Response Study!

Predictive Medicine, Inc. © 2010! 13!

Depression Background!

•  Clinical Depression!•  Treatment!•  Symptom Measurement!•  No specific diagnosis!•  No specific treatment!

Predictive Medicine, Inc. © 2010! 14!

Clinical Data! Symptoms!

! -HDRS (0-4 scale)!!

Treatment!-Desipramine (DMI)!-Cognitive Behavioral Therapy (CBT)!

!

Outcome!! - Responders!

Predictive Medicine, Inc. © 2010! 15!

Hamilton Psychiatric Scale for Depression!

Predictive Medicine, Inc. © 2010! 16!

Modelling !

!

Easier to understand!Easier to manipulate!Easier to analyze!

Recast  problem  into  mathematical  terms  

Predictive Medicine, Inc. © 2010! 17!

Predictive Medicine, Inc. © 2010! 18!

Understanding Recovery!

Predictive Medicine, Inc. © 2010! 19!

Depression Data!•  7 Symptoms ! !!

!Physical:! !E Sleep ! !! ! !M, L Sleep ! ! ! !! ! !Energy ! ! ! ! !!Performance: !Work & Interests ! ! ! !!Psychological: !Mood ! ! ! ! !! ! !Cognitions ! ! ! !! ! !Anxiety ! ! !!

•  2 Treatments ! !Cognitive Behavioural Therapy (CBT)! !! ! !Desipramine (DMI)!

!•  Clinical Data ! !Responders = improvement >= 50% ! !

! ! !N ! = 6 patient each study! ! !6 weeks ! = 252 data points each study!!

Predictive Medicine, Inc. © 2010! 20!

Overview Recovery Model and Parameters!

M

E W

MS

ES

A

C

Predictive Medicine, Inc. © 2010! 21!

Modeling Time to Response !

Predictive Medicine, Inc. © 2010! 22!

Modeling Treatment Effects!

Predictive Medicine, Inc. © 2010! 23!

Recovery Model Equation!

+

++

- ==

Predictive Medicine, Inc. © 2010! 24!

Training the model!

Predictive Medicine, Inc. © 2010! 25!

Recovery Pattern and ErrorExample Patient (CBT)!

Predictive Medicine, Inc. © 2010! 26!

Recovery Pattern and ErrorPatient Group (CBT)!

Predictive Medicine, Inc. © 2010! 27!

Overview!•  Why we did this work - to improve quality of life for millions

of people suffering from depression!•  How we did it - used differential equations (“neural network”)

to model and compare response to different antidepressant treatments!

•  What we found - different response patterns for the two treatments - the order and timing of improvement of symptoms were different!

•  What we think it means - improvement in selection of treatment - less trial and error !

Predictive Medicine, Inc. © 2010! 28!

ResultsOptimized parameters specify model

Initial conditions predict pattern trajectory !

M C

W A

E ES

MLS

Predictive Medicine, Inc. © 2010! 29!

Latency!

Predictive Medicine, Inc. © 2010! 30!

Mean ½ Reduction Time!

CBT varies 3.7 wks

DMI varies 1.8 wks

Predictive Medicine, Inc. © 2010! 31!

Direct Effect of Treatment!

Predictive Medicine, Inc. © 2010! 32!

Direct Treatment Intervention Effect!

Predictive Medicine, Inc. © 2010! 33!

Treatment Effects and Interactions!

CBT Sequential

DMI (delayed)

CONCURRENT

DMI: > 2x interactions and loops

Predictive Medicine, Inc. © 2010!

Order and Time of Symptoms Improve is Different for CBT and DMI!

Predictive Medicine, Inc. © 2010! 35!

Overview!•  Why we did this work - to improve quality of life for millions

of people suffering from depression!•  How we did it - used differential equations (“neural network”)

to model and compare response to different antidepressant treatments!

•  What we found - different response patterns for the two treatments - the order and timing of improvement of symptoms were different!

•  What we think it means - improvement in selection of treatment thereby reducing unnecessary costs and suffering. Potentially saving lives.!

Predictive Medicine, Inc. © 2010! 36!

Conclusions!•  An neural network model is capable of predicting

and describing recovery patterns in depression!•  We can do better than trial and error treatment

protocols!

•  Recovery patterns differ by treatment!•  Cognitive Behavioural Therapy!

is sequential!•  Desipramine!

is concurrent (after delay)!

•  Recovery patterns provide insights to patient response that can inform treatment choices!

Predictive Medicine, Inc. © 2010! 37!

Limitations!•  Model:!

•  Assumes symptoms interact!•  Assumes treatment acts directly!•  Permanent vs. transient!•  Causal vs. sequential!•  Statistical fluctuations not handled!

•  Study:!•  CBT measurement intervals vary!•  Small sample size!•  Initial 6 weeks of CBT (entire=16)!•  Finer resolution of measurements (2-3/day)!

Predictive Medicine, Inc. © 2010! 38!

Thank  you!  

Predictive Medicine, Inc. © 2010! 39!

Backup  Slides  

Predictive Medicine, Inc. © 2010!

Recovery Model!

Predictive Medicine, Inc. © 2010! 41!

Predictive Medicine, Inc. © 2010! 42!

Predictive Medicine, Inc. © 2010! 43!

Spanning  disciplines  Emerging  disciplines  

Diagnosis & Treatment

Depression Translational

Medicine

Huntington’s Disease

Medicine Life Sciences

Information Systems

Neuroscience Molecular Biology

Clinical Practice Signs and Symptoms

Biochemistry

Mathematical Models

Computer simulation

Ontology Genetics

Genomics

Anatomy

Machine Learning Semantic Web

Influenza

Research Findings

Bioinformatics Electronic Medical Records

Clinical Research Diabetes

Predictive Medicine, Inc. © 2010! 44!

Timeline  

2009 1993

World Congress on Neural Networks,

July 11-15, 1993, Portland,

Oregon

SIG Mental Function

and Dysfunction Sam Levin

Jackie Samson, Mc Lean Hospital

Depression Research

1996

1995

2008 1994

Patents Sold to Advanced

Biological Laboratories

Belgium

Patents Offered at

Ocean Tomo Auction

Chicago, IL

US Patent No. 6,317,73 Awarded

US Patents No.

6,063,028 Awarded

2001

2000

PhD Thesis

Proposal Approved

Workshop Neural

Modeling of Cognitive and Brain Disorders

BioPAX

? Linked Data W3C HCLS BioDASH

EPOS

2006

EMPWR

Poster Presented

ISMB 1997 PSB 1998

1997

Predictive Medicine, Inc. © 2010! 45! 27 October 2008

Neural  Modeling  of  Depression  1996 Luciano, J., Cohen, M. Samson, J. ”Neural Network Modeling of Unipolar Depression,” Neural Modeling of Cognitive and Brain Disorders, World Scientific Publishing Company, eds. J. Reggia and E. Ruppin and R. Berndt. Book cover; chapter pp 469-483.

Luciano Model highlighted on book cover

Workshop 1995 Book 1996

Predictive Medicine, Inc. © 2010! 46! 27 October 2008

 •  BioPathways  Consortium      •  BioPAX            •  W3C  Semantic  Web  for  Health  Care  and  Life  Sciences  (HCLSIG)  

Establishing  Communities  of  Interest/Practice  

Predictive Medicine, Inc. © 2010! 47! 27 October 2008

BioPAX  -­‐  Enabling  Cellular  Network  Process  Modeling  

Metabolic Pathways

Molecular Interaction Networks

Signaling Pathways

Gene Regulatory Networks

Glycolysis Protein-Protein Apoptosis TFs in E. coli

Predictive Medicine, Inc. © 2010! 48! 27 October 2008

EMPWR    Collaboration  with  Manchester,  UK  

• Use  instanceStore  to  reason  over  BioPAX  formatted  (OWL)  pathway  data  

• Goal:  discover  new  scientific  facts  • Method:  Utilize  power  of  reasoners  and  OWL  through  coupling  BioPAX  data  and  Manchester  Technology  • Results:  BioPAX  semantics  lacking  thus  had  to  educate  BioPAX  community  and  course-­‐correct  initiative  • Extending  BioPAX  to  enable  the  computational  exploration  

Predictive Medicine, Inc. © 2010! 49!

Diabetes  Type  2  !   90-­‐95%  diagnosed  cases  of  diabetes  (adults)  !   Usually  begins  as  insulin  resistance  !   Associated  with  age,  obesity,  family  history,  

history  of  gestatinal  diabestes,  impared  glucose  metabolish,  physical  inactivity,  race/ethnicity  

!   Rare  in  children,  but  increasing    

Predictive Medicine, Inc. © 2010! 50!

Understanding  the  role  of  risk  factors  in  insulin  resistance  

Figure: Integration of genomic and proteomic/metabolomic data (text boxes shaded in gray) proposed for current project. We hypothesize that diabetes risk factors result in altered gene and protein expression in skeletal muscle and adipose tissue (genomic data), leading to insulin resistance and inflammation. This, in turn, results in abnormal tissue function, as indicted by accumulation of long-chain fatty acyl CoA and oxidative damage (proteomic and metabolomic data), further insulin resistance and beta-cell failure, and ultimately to type 2 diabetes

Predictive Medicine, Inc. © 2010! 51!

Enhance  pathway  capability  !   Optimize    

!   Speed  

!   Accuracy  

!   Completeness  

!   Single  query  over  multiple  of  databases  

!   Validate,  test  and  evaluationr  !   Incorporate  into  diabetes  research  

workflow  

Predictive Medicine, Inc. © 2010! 52!

Enhance  pathway  capability  

Cell Designer model of adipose tissue cell. Add gene expression, standard metadata terms (BioPAX, GenBank)!Use with expression data constrained by proteomic data!towards target ID, biomarker ID, patient population ID!

Predictive Medicine, Inc. © 2010! 53! 27 October 2008

2008  Received  inquiry  and  put  up  for  auction  (Chicago)  2009  Sold  to  Advanced  Biomedical  Labs  (Belgium)  

US Patent No. 6,063,028 May 2000

AUTOMATED TREATMENT SELECTION METHOD

US Patent No. 6,317,731 Nov 2001

METHOD FOR PREDICTING THE OUTCOME OF A TREATMENT

OCEAN TOMO LLC Live Auction, Chicago, USA Oct 30, 2008

Expected Value $800,000+

Predictive Medicine, Inc. © 2010! 54!

Take  Home  Message    

•  We  need  to  shorten  the  time    • tighten  the  loop  between  research  and  practice;  

•  15  years  +  is  too  long,  way  too  long  

Predictive Medicine, Inc. © 2010! 55!

Acknowledgements  •  Sam  Levin  •  Dan  Levine  •  Dan  Bullock  •  Ennio  

Mingola  •  Michiro  

Negishi  •  Jacqueline  

Sampson  •  Larry  Hunter  •  Rick  Lathrop,  •  Larrie  Hutton  •  Tim  Clark    

Eric Neumann!Chris Sander!Mike Cary!Jeremy Zucker!Alan Ruttenberg!Jonathan Rees!Robert Stevens!Phil Lord!Alan Rector!Andy Brass!Paul Fisher!Carole Goble!George Church!Matt Temple!Christopher Brewster!

ME Patti!Mark Musen!Zak Kohane!Brian Athey!David States!!!!

Predictive Medicine, Inc. © 2010! 56!

Pre-­‐diabetes  

!   Increased  risk  of  developing  type  2  diabetes,  heart  disease,  and  stroke    

!   Blood  glucose  levels  higher  than  normal  (but  not  high  enough  to  be  characterized  as  diabetes)  

!   Impaired  fasting  glucose  (IFG),  impaired  glucose  tolerance  (IGT)  or  both.  

!   IFG  100  to  125  milligrams  per  deciliter  (mg/dL)  

!   IGT  140  to  199  mg/dL    

!   19%  adults  (US,  2007)  

!   7  %  IFG  adolescents  (US,  1999  to  2000)  

Source: http://diabetes.niddk.nih.gov/DM/PUBS/statistics/!

Predictive Medicine, Inc. © 2010! 57!

Diabetes  

http://diabetes.niddk.nih.gov/DM/PUBS/statistics/!

Predictive Medicine, Inc. © 2010! 58!

Data  !   130  nondiabetic  subjects  

!   Characterized  metabolically  

!   Family  history  pos  and  neg  

!   FH-­‐  more  insulin  sensitive  than  FH+    

!   Broad  range  of  insulin  sensitivity,  quartiles  of  SI  values  with  limits  of  2.6,  5.3,  and  8.4  

Subjects Recruited (May 2006)

(Mean + SD) Number 130 Age 36 + 10 years BMI 27 + 5 kg/m2 Gender 53 M, 77 F Family History DM 61 FH- neg, 69 FH+ pos Fasting Glucose 93 + 17 mg/dl Fasting Insulin 9 + 8 µU/ml SI 5.8 + 4.3 SG 0.0245 + 0.0211 AIRg 469 + 404

Predictive Medicine, Inc. © 2010! 59!

Research  Aims  

!   Enhance  diabetes  research  with  pathway  capability  for  target  identification,  biomarker  identification,  patient  population  identification  !   Enable  simulation  and  reasoning:  extend  and  

integrate  computational  technologies:  web  services,  workflows,  metadata,  ontologies  BioPAX  pathway  representation  

!   Optimize  the  speed,  accuracy  and  completeness:  single  query  over  multiple  of  databases.    

!   Deploy  into  diabetes  research  workflow  

Predictive Medicine, Inc. © 2010! 60!

Research  and  Practice  

!   Computational  modelers  construct  in  silico  representations  of  organic  phenomena    

!   Basic  researchers  construct  in  vitro  !   Clinical  Researcher’s  conduct  in  vivo  

studies  on  patient  populations  !   Clinical  practioners  apply  the  results  

of  clinical  research  

Predictive Medicine, Inc. © 2010! 61!

Questions!

•  Some people on antidepressants commit suicide. Is it possible that the antidepressant drug can cause this to happen?!

!•  How can differential equations help us

to understand what is going on?!