Contemporary philippine arts from the regions_PPT_Module_12 [Autosaved] (1).pptx
Cancer Research Meets Data Science — What Can We Do Together?
1. Cancer Research Meets Data
Science — What Can We Do
Together?
Phil Bourne, Lei Xie, Zheng Zhao & Cam Mura
School of Data Science & CUNY
November 19, 2021
2. Phil Bourne
• Dean & Prof School of Data
Science & Biomedical Eng. UVA
• Interests:
• Biomedical data sciences
• Scholarly communication
• Professional development
• Website:
• https://en.wikipedia.org/wiki/Philip
_Bourne
• http://bournelab.org/
• https://datascience.virginia.edu/peo
ple/phil-bourne
3. Agenda
• School of Data Science (SDS)
• Research opportunities
• Training
• Cancer relevant research
• Early stage drug discovery
• Multi-scale modeling
• The future – biomedical data sciences @ UVA
4. Agenda
• School of Data Science (SDS)
• Research opportunities
• Training
• Cancer relevant research
• Early stage drug discovery
• Multi-scale modeling
• The future – biomedical data sciences @ UVA
5. SDS Current Research Portfolio
12
7
4
3
2
3
3
Research Areas
Healthcare/Life Sciences
Technology/Software
Defense/Cybersecurity
Finance/Fintech
Energy/Environment
Education & Digital
Humanities
Faculty
• Phil Bourne
• Don Brown
• Tim Clark
• Teague Henry
• Jack van Horn
• Mike Porter
• Heman Shakeri
• Sana Syed
• Aidong Zhang
SDS strives to be a connector – a place where interdisciplinary
research driven by common data, methods and expertise
comes together
6. Sample of Biomedical Capstone Projects
• Improving analysis and reporting for Virginia's Division of Consolidated Laboratory Services newborn
screening program
• Analyzing national and state opioid abuse treatment completion with multilevel modeling
• Identifying risk of progression for patients with Chronic Kidney Disease using clustering models
• Predicting mortality risk associated with serious treatable surgical complications at the University of
Virginia health system
• Predicting patient revisits at the University of Virginia health system emergency department
• Priorities in hospital care across varying socioeconomic demographics
• Using machine learning to predict near-term mortality in cirrhosis patients hospitalized at the University
of Virginia health system
• Survival analysis of carbapenemase-producing enterobacteriaceae infections in hospital patients
• Optimization of sepsis risk assessment for ward patients
• Mixed linear modeling techniques for predicting fatalities in vehicle crashes
• Bi-directional relevance matching between medical corpora
• COVID 19 Health Disparities
• Prediction of decompensation in patients in the cardiac ward
• Geographic access to HIV care: The impact of ARRA
Touches basic and clinical care, patient wellness, health care efficiencies
7. Training
Researcher and Assistant Professor of
Medicine Dr. Thomas Hartka, also a
current online Masters in Data Science
student, is combining two disparate
data sets—electronic health records
and DMV crash data—to save lives
after motor vehicle crashes.
“I enrolled in the MSDS program to
expand my research on automotive
safety. I have already used
techniques from classes in my work.
I hope to expand my research to
real-time analytics to improve
emergency room care.”
— Dr. Thomas Hartka, UVA School
of Medicine
8. Training
•MSDS Online (2 year) and Residential (11
month) - 250 students
•Undergraduate minor - 170 students
•PhD program (next Academic Year) – Dual
mentorship ~ 40 students
•Undergraduate 4-year degree Academic Year
22-23
9. Agenda
•School of Data Science (SDS)
• Research opportunities
• Training
•Cancer relevant research
• Early stage drug discovery – systems pharmacology
• Using structural bioinformatics (Zheng)
• Using machine learning (Xie)
• Multi-scale modeling
• From molecules to patients (Mura)
•The future – biomedical data sciences @ UVA
10. Zheng Zhao
• Research Scientist in Dr. Bourne group
• Interests:
• Computational Drug Design and Discovery
• Biomedical Data Science Mining
• Website:
• http://bournelab.org/
• https://datascience.virginia.edu/people/zheng-zhao
• https://scholar.google.com/citations?user=siE8wmkAAAAJ&hl=en
11. Overall Approach - Data-driven targeted
(anti-cancer) drug screening
Given any interesting target:
validating target, exploring target,
drug screening, drug informatics,
SAR, etc.
Given any drug molecule:
potential (off) targets, drug-based
screening, binding characteristics,
molecular directed
modification/optimization, etc.
Potential drug molecules
Target-
based
screening
Drug-based
screening
Target-drug
interaction-
based
Zhao & Bourne. Journal of Proteome Research. 19 (11), 4698–4705, 2020.
12. Targeted drug design and discovery on a proteome
scale using a fingerprint-based approach
Phylogenetic kinome tree Potential kinase inhibitors
Proteome-scale Binding Characteristics
Protein-ligand interaction features
Zhao & Bourne, et.al Journal of Medicinal Chemistry 59 (9), 4326-4341, 2016
13. Predicting allosteric inhibitors
Exploring the features of
MEK-ligand interactions
Top 15 potential kinase targets
for allosteric inhibitor design
Predicting
Zhao & Bourne, et al. PLoS One 12 (6), e0179936, 2017
14. Designing covalent kinase inhibitors across the
human kinome
Zhao & Bourne, et. at. Journal of Medicinal Chemistry 60 (7), 2879–2889,
2017
Zhao & Bourne. Drug Discovery Today, V23 (3), 727-735, 2018
Zhao & Bourne. http://arXiv preprint arXiv:2106.11698,
2021
Reversible covalent kinase inhibitor
15. Exploring drug resistance mechanisms
EGFR structure dataset Mutations and conformations Change of the binding pockets for mutation
Zhao & Bourne, et al. Journal of Chemical Information and Modeling 59 (1), 453-462. 2019
16. Exploring drug (resistance) mechanisms
Molecular Dynamics
Umbrella sampling, Structural bioinformatics
Free energy calculations
Quantum mechanics/molecular mechanics approach
Background:
Crizotinib (first generation), Ceritinib, and Alectinib(Second generation) target ALK for treating NSCLC
Zhao & Bourne. J. Chem. Theory Comput. 16 (5), 3152-3161. 2020
17. Agenda
•School of Data Science (SDS)
• Research opportunities
• Training
•Cancer relevant research
• Early stage drug discovery – systems pharmacology
• Using structural bioinformatics (Zheng)
• Using machine learning (Xie)
• Multi-scale modeling
• From molecules to patients (Mura)
•The future – biomedical data sciences @ UVA
19. Deep Learning Powered Multi-scale
Predictive Modeling for Personalized
Anti-Cancer Drug Discovery
20. Predicting Genome-Wide Chemical-
Protein Interactions and Functional
Selectivity of Ligand Binding
• An integrated approach
• Structure-based data augmentation
• Self-supervised learning
• Meta-learning
https://pubs.acs.org/doi/full/10.1021/acs.jcim.0c01285
protein
sequence
Recepto
r
activity
antagonis
t
agonist
Not
bindin
g
Chemical
DeepReal
Bindin
g
Therapeutic
effect
Side effect
Biased
signaling
BioRxiv/2021/460001
21. Rational Discovery of Dual-Action Multi-
Target Anti-Cancer Therapy
• All anti-cancer therapies are associated with the increasing risk of heart failure
• Levosimendan, a drug for heat failure (target PDE3), is predicted to target RIOK1 gene
• KinomeScanTM and cell line assays validated the computational prediction.
Heart failure
Cancer
PDE3
Therapeutics
RIOK1
Multi-indication polypharmacology https://journals.plos.org/ploscompbiol/article?id=10.1371/journal.pcbi.1006619
22. Prediction of Chemical Phenomics of
Novel Cells and Patients for Mechanism-
driven Phenotype-based Compound
Screening
https://www.nature.com/articles/s42256-020-00285-9
https://doi.org/10.1101/2021.08.09.455708
+ +
MultiDCP
https://arxiv.org/pdf/2010.04824.pdf
https://doi.org/10.1101/2021.05.20.445055
Cancer cell line
CODE-AE
proteomics
transcriptomics
Cancer patients
23. Identifying Personalized Combination of
Immunotherapy and Targeted Therapy
• Many cancers do not respond to any anti-cancer therapies.
• Repurpose drugs to turn ”cold” tumors to “hot” ones for immunotherapy.
• PI3K/mTOR inhibitors can restore TET2 activity in aggressive squamous-like
pancreatic ductal adenocarcinoma subtype.
Repurposed
drug
‘Cold’ tumor
(TET2
inactive)
‘Hot’ tumor
(TET2
active)
Immunotherapy
Manuscript in revision
=
24. Cam Mura
• Senior Scientist, School of Data Science &
Biomedical Eng. UVA
• Interests:
• Biomolecular sequence, structure/dynamics,
function relationships, via ML through an
evolutionary lens
• What can EHRs teach us about the molecular basis
of disease?
• Mentoring: students and their research efforts
• Website:
• http://bournelab.org
• http://datascience.virginia.edu/people/cameron-
mura
25. Agenda
•School of Data Science (SDS)
• Research opportunities
• Training
•Cancer relevant research
• Early stage drug discovery – systems pharmacology
• Using structural bioinformatics (Zheng)
• Using machine learning (Xie)
• Multi-scale modeling
• From molecules to patients (Mura)
•The future – biomedical data sciences @ UVA
26. Motivating
question:
Can EHRs/RWE
teach us any-
thing about a
putative link
between
(some)
statins
and (some)
cancers?
Gohlke et al… Bourne, Stein & Preissner
https://www.medrxiv.org/content/10.1101/2021.07.20.21260891v1.full
Data pre-
processing and
cohort design
Multi-scale Modeling: Statins & Cancer
27. Age and gender distribution
in the trans-Atlantic cohort
and patients diagnosed with
cancer
The age distribution for both genders
in the full trans-Atlantic cohort is
shown (left half), along with the
analogous distribution for patients
diagnosed with cancer (right); any
effects that might stem from the
visible differences between these two
groups were mitigated by using a 1:1
matched study design.
Real World Evidence (RWE)
Gohlke et al… Bourne, Stein & Preissner
https://www.medrxiv.org/content/10.1101/2021.07.20.21260891v1.full
28. RWE, Statins & Cancer
Different cancer entities, and comparing different statins separately
Gohlke et al… Bourne, Stein & Preissner
https://www.medrxiv.org/content/10.1101/2021.07.20.21260891v1.full
29. RWE, Statins & Cancer
Statins (i) reduce MACC1
expression, (ii) specifically
inhibit MACC1-mediated
functions in vitro, and
(iii) decreases tumor burden
and metastasis formation
in vivo.
U Stein, W Walther, et al.
Max-Delbrück-Centrum für
Molekulare Medizin (MDC) Gohlke et al… Bourne, Stein & Preissner
https://www.medrxiv.org/content/10.1101/2021.07.20.21260891v1.full
30. RWE, Statins & Cancer
Statins (i) reduce MACC1
expression, (ii) specifically
inhibit MACC1-mediated
functions in vitro, and
(iii) decreases tumor burden
and metastasis formation
in vivo.
U Stein, W Walther, et al.
Max-Delbrück-Centrum für
Molekulare Medizin (MDC) Gohlke et al… Bourne, Stein & Preissner
https://www.medrxiv.org/content/10.1101/2021.07.20.21260891v1.full
31. RWE, Statins & Cancer: Conclusions so far
• Integrated model?—Considering MACC1 at the molecular & pathway level helps
elucidate possible statin ↭ cancer links by illuminating a potential network of
associations between aspirin, colorectal cancer, statins & MACC1, with HSPA5 as a
hub:
fat/glycogen/etc.
metabolism
(which is statin-related,
e.g. via the phenomenon
of statin muscle myalgia)
HSPA5
(a heat-shock
protein)
aspirin
colorectal cancer
MACC1
SH3BP4
(MACC1
paralog)
• Joint experimental & RWE-based analysis of statin ↭ cancer linkages.
Retrospective/observational, two-center (trans-Atlantic), nested case-control study of nearly
500K patient EHRs, sampled over ~10 y (+ further 132K statin-taking cancer patients in TriNetX)
• Experiments: Statins found in HTS; some can inhibit macc1 expt’n and ⇩ MACC1-
associated functions (e.g. motility, proliferation, metastasis in vivo)
• RWE: Various statins associate w/ 50% ⇩ overall risk of developing cancer; atorvastatin
exhibited greatest effects (OR 0.3; 95% CI 0.28-0.32). Most-strongly affected entities
were those of liver (OR 0.35), certain 2° neoplasms (OR 0.42), and CRC (OR 0.44).
Gohlke et al… Bourne, Stein & Preissner
https://www.medrxiv.org/content/10.1101/2021.07.20.21260891v1.full
32. Agenda
• School of Data Science (SDS)
• Research opportunities
• Training
• Cancer relevant research
• Early stage drug discovery – systems pharmacology
• Using structural bioinformatics (Zheng)
• Using machine learning (Xie)
• Multi-scale modeling
• From molecules to patients (Mura)
• The future – biomedical data sciences @ UVA
33. Biomedical Data Science - The Impact Will be
Profound
https://deepmind.com/blog/article/alphafold-a-solution-to-a-50-year-old-grand-challenge-in-biology
34. These QR codes link to each person’s scholar.google.com entry
36. Examples of our interests [rough/draft]
• In silico, early-stage drug discovery &
repurposing, utilizing EHRs and RWE
• Multi-scale modeling: from molecules
to populations (& back to molecules)
• Generally rely upon structure-based
approaches (structural bioinformatics)
• Health disparities research (recent)
37. Our philosophy: Let data guide our models
Mura, Draizen, Bourne. "Structural Biology Meets Data Science: Does
Anything Change?"; COSB (2018) ♢ DOI: 10.1016/j.sbi.2018.09.003
We’re interested in human health!
…and our roots—and many ongoing
projects—are in basic research
The data (volumes, and types) &
methodologies (DL) now available
enable unprecedented questions
to be pursued (e.g., AlphaFold)
Rough draft for now. Unsure if we should keep this
slide? — I like it mainly as a segue to molecules/
Urfold (if we end up including that), in which case it
could come later, right before the Urfold section.
38. Towards data-guided multiscale modeling
Sejnowski, "The Unreasonable Effectiveness of Deep Learning in AI"; PNAS (2020)
…from
molecules ⤏⤏
organisms ⤏⤏
populations of
organisms
40. Approach, quantitative: DeepUrfold
Jaiswal… Draizen… Mura, Bourne. "Deep Learning of Protein Structural Classes: Any Evidence for an ‘Urfold’?“
2020 Systems & Info Eng Design Symp (IEEE SIEDS Proceedings) ♢ DOI: 10.1109/SIEDS49339.2020.9106642
Hinweis der Redaktion
I will introduce the concept of data science with a story that illustrates - citizen engagement, merging of unexpected data and societal benefit