Artificial Intelligence in Translational Precision Medicine - - PowerPoint PPT Presentation
Artificial Intelligence in Translational Precision Medicine ACOSIS-2019 Marrakech, Morocco Nov 20-22 nd , 2019 Peter J. Tonellato, PhD Professor of Bioinformatics Director of Center for Biomedical Informatics Health management and
ACOSIS-2019 Marrakech, Morocco Nov 20-22nd, 2019
Professor of Bioinformatics Director of Center for Biomedical Informatics Health management and Informatics School of Medicine, University of Missouri Columbia, Missouri, USA
NIH and US academic healthcare complexes have turned attention to data intensive, evidence-based, patient centric “Precision Medicine” – accounting for individual patient genetics, lifestyle and environment. Era of digitized patho/physiology - “big” data using emerging digital sequencers; high definition 3/4-D imaging, … Seek a translational approach capable of restoring personalized medicine while leveraging ‘big’ data and analytics with objectives:
Accelerate De-Personalization with Data-Driven Medicine
and information at Terabyte levels
risks far beyond technical weaknesses in approach, methods, sensitivity and specificity
healthcare providers
driving de-personalization Uber De-Personalized
Clinical Enterprise Research Enterprise
Translation
PHI First Name: Ozzy Last Name: Osborne Physical Height: 6’ Weight: 160 Genetic CYP2C9: *1/*1 VKORC1: A/A PHI First Name: Animal Last Name: House Physical Height: 6’ 6” Weight: 180 Genetic CYP2C9: *3/*3 VKORC1: A/B
Phenomenological modeling provides iterative method to accurate representations
Ken from Toy Story 3
Variable(s) Parameters Age 18 to 24 (21.1%), 25 to 44 (30.3%), 45 to 64 (21.9%), 65 to 94 (26.7%) Gender Male {< 18 (51.26%), 18 to 24 (51.11%), 25 to 44 (50.06%), 45 to 64 (48.65%), 65 and
(51.35%), 65 and over (58.82%)} Race White (75.1%), African American (12.3%), Native American (0.9%), Asian (3.6%), Pacific Islander (0.1%), Other (5.5%), Unknown (2.5%) Height Mean: 69.2”, St.D: 6.6”, Min : 56.0”, Max: 82.4” Weight Mean: 189.8 lb, St.D: 59.1 lb, Min: 71.6 lb, Max : 308.0 lb Smoker White - 20%; African American - 21%; Native American - 35%; Asian / Pacific Islander - 11%; Other - 23% Amiodarone Y - 55%, N - 45% DVT Y - 26.8% N - 73.2% VKORC1 A/A - 65%, A/B - 20%, B/B - 15% CYP2C9 *1/*1 - 64.3%, *1/*2 - 18%, *1/*3 - 11.7% , *2/*2 - 2% , *2/*3 - 2.1% , *3/*3 - 0.25% The clinical avatar population and the resulting variables and statistical distributions.
“Optimal” BNM.
Knowledgei Knowledge2
Data
Imputation
Train Data
Random Sampling ~70%
PC JCPC JPC
FCI FCI
PCL GES CPC
BNMs
BN2 * BN2i * BN1 * BN1i *
Knowledge1
Validation Markov Blankets
Test Data Test Data
Testing/Predictive Metrics
Test Data
Validation Literature/Experts
BN3 * BN3i *
21
DAG 1 DAG 5 DAG 2 DAG 3 DAG 4 DAG 6
Parameter U.S. Base Actual PharmGKB Warfarin18 (n=5700) Simulated Warfarin (n = 20,000) P-Value* Age1 <18 18 – 24 25 – 44 45 – 64 65 – 94 27.6% (1572) 7.4% (420) 30.9% (1763) 21.5% (1227) 2.6% (718) 0.18% (10) 1.3% (75) 9.9% (559) 36% (2,040) 52.5% (2,974) 0.13% (26) 1.2% (235) 9.8% (1,957) 36.4% (7,282) 52.5% (10,500) 0.75
Gender by age1 <18 18 – 24 25 – 44 45 – 64 65 – 94 M: 49.9% (784), F: 50.1% (788) M: 48.8% (205), F: 51.2% (215) M: 50% (882), F: 50% (881) M: 48.4% (594), F: 51.6% (633) M: 41.4% (310), F: 58.6% (438) M: 30% (3), F: 70% (7) M: 42.7% (32), F: 57.3% (43) M: 49.9% (279), F: 50.1% (280) M: 60% (1225), F: 40% (815) M: 59.3% (1,855), F: 40.7% (1,272) M: 34.6% (9), F: 65.4% (17) M: 47.2% (111), F: 52.7% (124) M: 50.6% (990), F: 49.4% (967) M: 59.4% (4,324), F: 40.6% (2,958) M: 59.4% (6,353), F: 40.6% (4,344) 0.89 Race1 White African American Native American Asian Pacific Islander Other Unknown 75.1% (4,282) 12% (684) 0.8% (45) 3.6% (206) 0.09% (5) 5.9% (336) 2.5% (142) 54.8% (3,122) 8.1% (462) 0% (0) 28.7% (1,634) 0% (0) 0% (0) 8.4% (482) 54.2% (10,835) 7.9% (1,583) 0% 29.7% (5,936) 0% 0% 8.2% (1,646) 0.51
Parameter U.S. Base Actual PharmGKB Warfarin18 (n=5700) Simulated Warfarin (n = 20,000) P-Value*
Weight5 (lbs) Mean Min Max 176.55 ± 30.9 92 273 171.58 ± 48.2 66 524 173.51 ± 27.87 92 290 7.7e-31 Smoker4 White African American Native American Asian/Pacific Islander Other/Unknown 20.3% (868) 21.2% (145) 40% (18) 11.7% (24) 22.6% (76) 14.4% (324) 20.8% (91) 0% (0) 6.4% (18) 6.5% (16) 14.3% (1,552) 20.9% (332) 0% (0) 5.7% (340) 5.7% (94) 2.4e-11 DVT6,7,8,9,19,11 Yes No 26.8% (1,527) 73.2% (4,173) 16.4% (817) 83.6% (4,191) 16% (3,203) 84% (16,797) VKORC17,10,14,15,16,17 A/A A/B B/B 15.5% (884) 46.7% (2,661) 37.8% (2,155) 52.2% (1,245) 25.8% (614) 22.0% (525) 52% (10,404) 26.3% (5,261) 21.7% (4,335) 0.83 CYP2C97,10,14,15,16 *1/*1 *1/*2 *1/*3 *2/*2 *2/*3 *3/*3 64.3% (3,666) 18.8% (1,071) 12.6% (718) 1.9% (109) 2% (114) 0.39% (22) 74.9% (4,155) 13.4% (742) 9% (501) 1% (58) 1.3% (72) 0.4% (22) 75.4% (15, 079) 13.4% (2,676) 8.8% (1,756) 1% (194) 1.1% (227) 0.3% (68) 0.81
Use Avatar Data to Predict Therapeutic Dose Statistical Characterization Analysis and Interpretation of Avatar Population Phenomenological Stochastic Model
Analyze Results
Interpretation of Simulation
Instantiate Clinical Avatar Population Adjustment of the Model
Simulate Milwaukee County*
(n=930,261)
and Cairo#
(n=7,137,218)
Analyze Results
*Wisconsin Interactive Statistics on Health (WISH). Wisconsin Department of Health Services; 2009.
#Egypt Demographic and Health Survey; 2008. World Health Organization (WHO); 2007.
Predict Initial Dose
Simulate populations based on characteristics such as genotype, age, and race
(0.0075*AGE) - (0.20*CYP2C92) + (0.20*TINR) + (0.09*SMOKER) - (0.09*RACE) + (0.07*DVT) - (0.25*AMI))
+ WEIGHT*(0.003)))/7 2-compartment model with 1st
2007]
AD 1 2
K21 K12 K10 Ka 1. Coumagen trial 2. Wilson, 2007 3. Fixed percent adjustment Time in therapeutic range (TTR)
1 2 3 4 5 6 7 8 9 10 Days 2 3 Time INR
Days 1 - 2 10 mg daily dose Days 8 - 90 Days 3 - 7
Group Percent TTR In-Range
20 40 60 80 Original Coumagen PGx STD Coumagen Simulation PGx STD Group PGx STD
Anderson, Circulation, 2007
P = 0.47 P = 0.48
Coumagen P = 0.48 Wilson P = 0.005
Protocol2 Protocol3 Protocol4 Protocol5 Protocol1 Protocol* with the highest balance of benefit, risks and, costs
Patient sub-population characteristics Caucasian, CYP2C9=*1/*1, and VKORC1=A/A
Benefit 1
Protocol6
Risks 1 Costs 1 Benefit 2 Risks 2 Costs 2 Benefit 3 Risks 3 Costs 3 Benefit 4 Risks 4 Costs 4 Benefit 5 Risks 5 Costs 5 Benefit 6 Risks 6 Costs 6 Stroke prevention (benefit) Bleeding (risks) X X X X X X
Optimal region Example
Time in therapeutic range
The optimal protocols significantly improve TTR by 16% and 12% (1.8 and 1.3 fewer adverse events) from uniform pharmacogenetic protocol for the rare and hard to manage patient subgroup
Center for Biomedical Informatics School of Medicine, University of Missouri, Columbia, Missouri Overall objective
First TPMDCP Tunisia- US Event
TPMDCP objectives and Description of Work
Pasteur Institute of Tunis
Venoms & Therapeutic Molecules lab (LR 16- IPT 08) NanoBioMedika Reasearch Team
University of Missouri
Workpackage WP 1. Comparative Analysis Tunisien subjects recruitment and data generated Identify new biomarkers useful for cancer target therapy, using patient’s phenotypes to predict gene mutations important to the more individualized treatment in Tunisia data already generated Identify new biomarkers useful for cancer target therapy, using patient’s phenotypes to predict gene mutations important to the more individualized treatment in Tunisia data already generated Specific Objective 1. Tunisian DCP Collect U.S subjects, generate data and biobank construction The same study design, approach and bioinformatics methods and analysis will be used to conduct the study both in the US and in Tunisia. A Genome Wide Association Study (GWAS) will be conducted to identify the biomarkers that predict the gene mutations and outcomes, in both populations Specific Objective 2. United States DCP To compare results of Tunisia and USA studies and to find both similar and divergent risk factors, biomarkers and functional implications among the two populations Workpackage WP2. Omics Computation Training To establish an international training program on biomedical informatics. During this collaborative study period, both Tunisia and United State project team members will be trained in research approach, methods and computational skills for multi-omics analysis First TPMDCP Workshop on Omics-Computing Precision Medicine in the Era of “Big Data” Genomics, December 19th, 2018
Identify novel molecular/genetic mechanisms associated with Digestive Cancers (DC) Identify novel molecular/genetic mechanisms associated with Digestive Cancers (DC)
Develop test biomarkers from novel DC molecular genetics mechanism Develop test biomarkers from novel DC molecular genetics mechanism
Implement the DC biomarkers for diagnostic, prognostics and therapeutic efficacy for DC Implement the DC biomarkers for diagnostic, prognostics and therapeutic efficacy for DC
Tunisien subjects recruitment and data generated Identify new biomarkers useful for cancer target therapy, using patient’s phenotypes to predict gene mutations important to the more individualized treatment in Tunisia data already generated Specific Objective 1. Tunisian DCP 2
Case-Control Study Case-Control Study
300 Controls
110 RC 54 56 190 CC 92 98 91 209 7 3
310 Cases (DC)
Study Design/ Subjects
10 GC
Large Pedigree Study Large Pedigree Study
12 with DC 38 Members 11 Subjects
3 Cases 8 Controls 12
Clinical and Laboratory Variables
Clinical data
BMI Gender Age AHT Anemia Glycemia Patient’s cancer background
Lifestyle
Vegetable consumption Meat consumption Brine consumption Fat consumption Smoking Alcohol
Tumor information
Histological type Cell differentiation Tumo size Tumor stage Tumor Site
Biochemical and Hematologic tests Kinetics variation of tumor markers Treatment response
Totol Protein Hematology Hemoglobine Leucocytes Platelet Reticlocytes Ca19.9 CEA Respander Not-Respander Tretment Tolirence
Treatment Protocols
Surgery Chmioterapy Radyotherapy Targeted Terapy
16
Conclusion Conclusion
32
Sample Collection Sequencing Analysis Clinical Action
Instances Storage
AWS
Variant calling Annotation Preparation/Alignm ent
GenomeKey
Workflow management System Job splitting Web Interface Job tracking
Grid engine Gluster FS MySQL DB
Networking
Complex Neurologic Diseases Complex Cardiovascular Diseases
Edward (Ted) H. Shortliffe
Perhaps first advocate after his BS in Math (Harvard) and MD and PhD (Stanford, 1975, 1976) with dissertation on MYCIN system * – rule-based AI CDS (clinical decision support) to diagnosis source and recommend treatment of infection. Not used in practice but validation work demonstrated accuracy arguably better than that of infectious disease physicians.
Shortliffe also founded the field and defined Biomedical informatics (BMI) as the interdisciplinary field that studies and pursues the effective uses of biomedical data, information, and knowledge for scientific inquiry, problem solving, and decision making, driven by efforts to improve human health.**
*Shortliffe, E.H.; Buchanan, B.G. (1975). "A model of inexact reasoning in medicine". Mathematical Biosciences. 23 (3–4): 351–379. doi:10.1016/0025- 5564(75)90047-4. MR 0381762 (online: http://www.shortliffe.net/Buchanan-Shortliffe-1984/MYCIN%20Book.htm, in particular, chapter 5). **Kulikowski CA, Shortliffe EH, Currie LM, Elkin PL, Hunter LE, Johnson TR, Kalet IJ, Lenert LA, Musen MA, Ozbolt JG, et al. AMIA board white paper: definition of biomedical informatics and specification of core competencies for graduate education in the discipline. J Am Med Inform Assoc. 2012;19(6):931–938. doi: 10.1136/amiajnl-2012-001053.
https://towardsdatascience.com/artificial-intelligence-framework-a-visual-introduction-to-machine- learning-and-ai-d7e36b304f87
NIH Aug, 2018 meeting to develop a roadmap for future AI in imaging research initiatives:
Research priorities highlighted in the report include:
interpretation from source data,
the imaging report, electronic phenotyping, and prospective structured image reporting,
architectures, and distributed machine learning methods,
called explainable artificial intelligence), and
Langlotz, C., Allen, B., Erickson, B., Kalpathy-Cramer, J., Bigelow, K., Cook, T., Flanders, A., Lungren, M., Mendelson, D., Rudie, J., Wang,
NIH/RSNA/ACR/The Academy Workshop. Radiology, 291(3), pp.781-791.
Langlotz, C., Allen, B., Erickson, B., Kalpathy-Cramer, J., Bigelow, K., Cook, T., Flanders, A., Lungren, M., Mendelson, D., Rudie, J., Wang,
NIH/RSNA/ACR/The Academy Workshop. Radiology, 291(3), pp.781-791.
Alphabet, Inc acquires DeepMind Technologies (2014) leads to development of AlphaStar (2019, almost realtime learning of multiplayer player games – chess, shogi, Go, StarCraft II, … mainly through self-play learning) running on Edge TPU (Tensor Processing Unit) a proprietary AI accelerator application-specific integrated circuit (ASIC) specifically designed for google’s TensorFlow framework. Google Health has taken over healthcare applications of DeepMind origin technologies (blog.google/technology/health) TensorFlow: www.tensorflow.org Versions available for CPU, GPU amd for google’s cloud available TPU. TensorFlow2 on macOS (www.tensorflow.org/install) Or test on google colab: www.tensorflow.org/tutorials With accompanying datasets – e.g. Human Variant Annotation Datasets:
https://console.cloud.google.com/marketplace/details/bigquery-public-data/human-variant-annotation-public?filter=solution- type:dataset&filter=category:genomics&id=9e418c65-7c29-471f-8539-9557e96f807c
ImageNet Large-Scale Visual Recognition Challenge (ILSVRC). Error rates dramatically improved with the introduction of deep learning in 2012. Human error rate continues at approximately 5%. General and application-specific challenges rapidly emerging.
Russakovsky, O., Deng, J., Su, H. et al. Int J Comput Vis (2015) 115: 211. https://doi.org/10.1007/s11263- 015-0816-y image-net.org/challenges/LSVRC/ Current status: www.forbes.com/sites/aarontilley/2017/07/31/china-ai- imagenet/#19b5f1fe170a
American College of Radiology Use Cases: www.acrdsi.org/DSI-Services/Define-AI
Abdominal: Detect acute appendicitis: Detect polyps greater than six millimeters Breast: Classifying Suspicious Microcalcifications Breast Density Quantification Cardiac: Aortic Valve Size Ascending Aortic Diameter Cardiac Output (LV Assessment) Cardiomegaly (size) Cardiothoracic Ratio (size) Carina Angle Measurement Coronary Flow Reserve (CFR) Flow in Ascending Aorta Flow in Pulmonary Artery Left Atrial Enlargement Left Atrial Size LV Late Gadolinium Enhancement Left Ventricle Myocardial Mass Left Ventricle T1 Mapping Quantification Left Ventricle Volume Left Ventricle Wall Motion Left Ventricle Wall Thickening Left Ventricle Wall Thickness Measurement Left Ventricular Late Gadolinium Enhancement Myocardial Perfusion Quantification for CT Pulmonary Artery Diameter Pulmonary Artery to Aortic Diameter Ratio Pulmonary to Systemic Flow Ratio Pulmonary Veins Mapping Preablation TAVR Aortic Root Measurements Musculoskeletal: Accessory Muscles in Neurovascular Compromise Chrondral Bone Lesion Characterization Hip Osteolysis Hip Subsidence Ligamentum Teres Injury Detection Neurology: Midline Shift Motor Cortex Quantitative Mapping Oncology: Lymph node involvement and extranodal extension Pediatric: Scoliosis Thoracic: Incidental Pulmonary Nodules on Radiographs Incidental Pulmonary Nodules on CT Pneumonia Pneumothorax Tuberculosis Screening Tuberculosis Triage
Shen, L., Margolies, L.R., Rothstein, J.H. et al. Deep Learning to Improve Breast Cancer Detection on Screening Mammography. Sci Rep 9, 12495 (2019) doi:10.1038/s41598-019- 48995-4
Conclusions The AI system (Transpara 14.0, Screenpoint Medical) achieved a cancer detection accuracy comparable to an average breast radiologist in this retrospective setting. Although promising, the performance and impact of such a system in a screening setting needs further investigation.
Alejandro Rodriguez-Ruiz, Kristina Lång, … Ioannis Sechopoulos, Stand-Alone Artificial Intelligence for Breast Cancer Detection in Mammography: Comparison With 101 Radiologists, JNCI: Journal of the National Cancer Institute, Volume 111, Issue 9, September 2019, Pages 916–922, https://doi.org/10.1093/jnci/djy222
Almira, A. Bagherzadeh, M, Akbari, M.R., Liquid biopsy in breast cancer: A comprehensive review , Clinical Genetics, Volume: 95, Issue: 6, Pages: 643-660, First published: 22 January 2019, DOI: (10.1111/cge.13514) Wan, N., Weinberg, D., Liu, T. et al. Machine learning enables detection of early-stage colorectal cancer by whole-genome sequencing of plasma cell-free DNA. BMC Cancer 19, 832 (2019) doi:10.1186/s12885-019-6003-8
American College of Radiology Use Cases: www.acrdsi.org/DSI-Services/Define-AI
Abdominal: Detect acute appendicitis: Detect polyps greater than six millimeters Breast: Classifying Suspicious Microcalcifications Breast Density Quantification Cardiac: Aortic Valve Size Ascending Aortic Diameter Cardiac Output (LV Assessment) Cardiomegaly (size) Cardiothoracic Ratio (size) Carina Angle Measurement Coronary Flow Reserve (CFR) Flow in Ascending Aorta Flow in Pulmonary Artery Left Atrial Enlargement Left Atrial Size LV Late Gadolinium Enhancement Left Ventricle Myocardial Mass Left Ventricle T1 Mapping Quantification Left Ventricle Volume Left Ventricle Wall Motion Left Ventricle Wall Thickening Left Ventricle Wall Thickness Measurement Left Ventricular Late Gadolinium Enhancement Myocardial Perfusion Quantification for CT Pulmonary Artery Diameter Pulmonary Artery to Aortic Diameter Ratio Pulmonary to Systemic Flow Ratio Pulmonary Veins Mapping Preablation TAVR Aortic Root Measurements Musculoskeletal: Accessory Muscles in Neurovascular Compromise Chrondral Bone Lesion Characterization Hip Osteolysis Hip Subsidence Ligamentum Teres Injury Detection Neurology: Midline Shift Motor Cortex Quantitative Mapping Oncology: Lymph node involvement and extranodal extension Pediatric: Scoliosis Thoracic: Incidental Pulmonary Nodules on Radiographs Incidental Pulmonary Nodules on CT Pneumonia Pneumothorax Tuberculosis Screening Tuberculosis Triage
Marrakech, Morocco Nov 20-22nd, 2019
Professor of Bioinformatics Director of Center for Biomedical Informatics Health management and Informatics School of Medicine, University of Missouri Columbia, Missouri, USA
Normal/Peritumoral tissues Tumor tissues
Picture : Web source
PBMC tissues
Tumor, Normal and PBMC
N Stage Normal Tumor PBMC Total 3 Nocancer (+1) (+1) (+1) 3 27 Stage 1 22 (+1) 22(+2) 6(+2) 55 10 Stage 2 9 9(+1) 1 20 6 Stage 3 4 4 (+2) 10 4 Stage 4 1 1(+1) (+2) 5
1 No NSCLC- Mesothelioma 1 1 2
Total 51 95 NSCLC: tumor/ normal lung tissue pairs : 29 (125, 14, 51, 18, 23, 2, 31, 34, 35, 4, 52, 54, 55, 56, 60, 62, 63, 64, 65, 68, 69, 6, 70, 77, 8, VA2, VA3, VA5, 17) NSCLC: tumor/normal lung tissue pairs + PBMC: 7 (119, 71, 72, 75, 78, 81, 95) NSCLC tissue only : 5 (5, 12, 32, 33, 57) NSCLC PBMC only : 6 (73, 74, 93, 94, 101, 103) Mesothelioma tumor/normal pair : 1 pair (21) Benign pathologic tissue sample : 1 (10T) Benign normal lung tissue : 1 (13) Benign PBMC : 1 (76) TOTAL 95 SAMPLES, 51 PATIENTS Stage Grade Total No Cancer 3 1 55 1 5 2 37 3 13 2 20 2 12 3 8 3 10 2 5 3 5 4 5 1 1 2 3 4 1 No NSCLC_ Mesothelioma 2 Grand Total 95
Stage & Grade Stage 1: 22
1: 2 2: 15 3: 5 Stage 2: 9
2: 5 3: 4 Stage 3: 4
2: 2 3: 2 Stage 4: 1
2: 1
https://www.broadinstitute.org